The family law court held a July 30, 2025, hearing on a motion to modify timesharing. The court wrote:

During this [July 30^th] hearing, Appellant [Ms. Adams] testified about taking one of the children to an urgent care center and receiving medication after the appointment. She also produced a picture of the prescription medication bottle which was entered into evidence.

On August 13, 2025, Appellee [Mr. Anderson] filed a motion for sanctions alleging that Appellant’s testimony about the urgent care visit was false and that the picture of the medication had been altered or was created by artificial intelligence. Appellee’s motion stated that he had requested medical records from the urgent care center and the pharmacy, but both claimed no records existed for the time the child was said to have been taken to the urgent care center.

In response to the motion, Appellant denied the allegations and provided more documents to support the child’s going to the urgent care center and receiving medication. Appellee’s reply to the response indicated that the additional documents were also fake, photoshopped, or created by artificial intelligence.

Id. at *1 (emphasis added).

Initially, it is worth noting that Mr. Anderson’s challenge was more than a mere accusation. It was supported by evidence. Specifically, “he had requested medical records from the urgent care center and the pharmacy, but both claimed no records existed for the time the child was said to have been taken to the urgent care center.”  His challenge was supported by a factual predicate.

Several weeks later, on August 22^nd, the court addressed the issue. “The court indicated that, if the documents were falsified and her testimony untrue, Appellant would be going to jail.”  Id. at *1.

What happened next was crucial: “Counsel for Appellant indicated that Appellant’s testimony and the documents submitted to the court were false or altered.”  Id.  The court then issued a show cause order.

After some procedural steps, Ms. Adams entered a guilty plea to the contempt.

The court then swore in Appellant and questioned her about her testimony during the July hearing and the fake and altered documents entered into evidence. Appellant admitted to lying to the court. After a brief recess, Appellant then apologized to the court and the court sentenced her to forty-five days in jail. This appeal followed.

The court addressed a number of issues unrelated to the deepfakes, such as mootness, separation of powers, procedures for taking a guilty plea, and whether issues were preserved for appeal.  As to due process, the court wrote:

Here, we believe Appellant’s false testimony to the court and fabricated documentary evidence were indirect criminal contempt. We believe so because it took additional evidence from Appellee to bring Appellant’s actions to light. Due to this being indirect criminal contempt, the court had to provide more than a summary adjudication. Considering the case law cited above and the fact that we are reviewing for palpable error, we conclude that no error, palpable or otherwise, occurred as it pertains to due process.

Id. at *3 (emphasis added).  As to contempt, it added:

While it is true that Appellant did not disobey an order of the court, she did obstruct justice and disrespect the court. Providing false testimony during a hearing and submitting altered or fake documents into evidence is the epitome of resisting the authority of the court and degrading its authority. Furthermore, the court did not try to convict Appellant of a crime; it simply held her accountable for lying to the court. The court did not err in holding Appellant in contempt.

Id. at *4 (emphasis added).  Addressing separation of powers, the court wrote:

Appellant’s next argument is that the court violated the separation of powers doctrine by charging her with the crimes of perjury and forgery. This argument is without merit. As previously stated, the court held Appellant in contempt for lying to the court and introducing fake documents into the record. The judicial branch has the power to vindicate its own authority without complete dependence on the executive and legislative branches…. The court did not charge Appellant with or convict Appellant of a crime; therefore, there was no violation of the separation of powers.

Id. at *4 (emphasis added; cleaned up).

The court concluded: “Appellant’s actions were gravely serious and undermined the administration of justice. Pursuant to our palpable error review of this case, we find no manifest injustice and affirm the judgment of the trial court.” Id. at *5.

The Adams court wrote, twice, that the fake photos were entered into evidence. It wrote once that  she introduced fake documents “into the record.”

When Ms. Adams offered the initial photo, Mr. Anderson apparently did not have the evidence from the health care provider to challenge authenticity.  The first hearing was on July 30^th and the health care provider’s evidence was submitted as part of a sanctions motion on August 13^th.  This illustrates the need for early discovery and preparation to challenge evidence on the basis that it is a deepfake.

This blog was initially posted on  Electronic Discovery Reference Model and  JD Supra.

In Rubesne Resources LLC v. ACE Property and Cas. Ins. Co,. 2026 WL 1251605, at *5 (D. Col. May 7, 2026), “ACE’s principal request in the Motion to Compel is that the court order a forensic examination of Ms. Rubesne’s cell phone by an expert to be identified and retained by ACE, who would be tasked with creating a mirror image of the device in order to attempt to recover any deleted social media and text messages within the circumscribed timeframe of December 1, 2023, to January 7, 2024, and to identify any deletion activity concerning this narrow group of texts.

The court wrote:

To begin with the governing legal principles, “[w]hen determining whether to grant a motion to compel the forensic imaging of a cell phone or other electronic device, courts have considered whether the examination will reveal information that is relevant to the claims and defenses in the pending matter and whether such an examination is proportional to the needs of the case given the cell phone owner’s compelling privacy interest in the contents of his or her cell phone.” … The court may compel a forensic examination of a personal device when the moving party demonstrates that the opposing party has concealed information or lacks the expertise necessary to search and retrieve all relevant data. … “Further, before permitting an intrusion into an opposing party’s information system—particularly where that party has undertaken its own search and forensic analysis and has sworn to its accuracy⁵ —the inquiring party must present at least some reliable information that the opposing party’s representations are misleading or substantively inaccurate.”…

Id. at *5.

On the facts presented: “The court finds that ACE has come forward with sufficient reliable information indicating that Rubesne’s representations concerning the Missing Texts are misleading, if not inaccurate, and that the nebulous circumstances surrounding Rubesne’s own forensic examination call into question whether all avenues have been exhausted to search and retrieve potentially available information from Ms. Rubesne’s cellular phone for examination. Additionally, the court concludes that no other legal barrier precludes the issuance of an order directing Rubesne to produce the device. In so finding, the court emphasizes four key points.”  Id.

One of many points in the court’s comprehensive analysis was:

[T]he first forensic examination completed by Rubesne, concerning which no particulars have been revealed, was undertaken without notice to ACE. Had Rubesne immediately owned up to the Missing Texts, and conferred with ACE before having Ms. Rubesne’s phone examined, the instant dispute may have been averted. The parties might have agreed on a neutral third-party forensic expert, as well as the parameters of a forensic examination. As things stand, however, Rubesne has conducted its own forensic examination, but would have the court preclude ACE from conducting a similar analysis—leaving ACE (and the court) to take Rubesne’s word for it that there is no other responsive, non-privileged information to be extracted from Ms. Rubesne’s phone. In light of the record described above, the court perceives no justification for countenancing this unequal approach and finds that ACE is entitled to test the integrity of Rubesne’s forensic examination by conducting one of its own.

Id. (emphasis added).

See generally:

• Is There a Duty to “’Fess Up?” – Part II
• Is There a Duty to “’Fess Up?”
• Candor Pays Off When It Comes to Alleged Spoliation: Notice + Accrual = Claims Bar

Burnley involved two motions to enforce settlement agreements.

Mr. Burnley had made harassment claims against his neighbors and the police.  The settlement agreements were memorialized in two separate writings, and the case was dismissed, with the court retaining jurisdiction to enforce the agreement.  A party seeking to enforce the agreement was first required to notify the opponent, “describing the alleged breach and supplying written, audio, video, photographic, or other tangible evidence of the alleged breach.”

Later: “Walburn provided Burnley with a written notice of breach…. Specifically, counsel for Walburn sent, by USPS Priority Mail, a letter to Burnley identifying the provision of the Settlement Agreement he allegedly breached, providing a copy of the Settlement Agreement, and enclosing the audio recording of Burnley allegedly calling Walburn’s employer.” [emphasis added].

Both settling parties moved to enforce the settlements.  Mr. Burnley filed responses.  After holding that the agreements were valid, the court wrote:

The question remains as to whether Burnley breached these settlement agreements. For the reasons stated below, the Court finds that Burnley violated the Walburn-Burnley Settlement Agreement by contacting Walburn’s employer about Walburn.

It then turned to the facts.  A caller, identifying himself as “Tyrone Jackson,” began making calls to Mr. Walburn’s employer.  Suffice it to say that the calls breached the agreement if they were from Mr. Burnley. Mr. Walburn contended that they were and that “Tyrone Jackson” was a pseudonym.  He pointed to a number of facts, such as the fact that “Tyrone Jackson” was a former neighbor of Mr. Burnley. Mr. Burnley denied making the calls and challenged voice comparison. The court wrote:

He argues that the Court cannot grant the Motion to Enforce because the audio recordings have not been “authenticated prior to admission” pursuant to Federal Rule of Evidence 901(a)…. Instead, he argues, the recordings are either someone else entirely or are a manipulated audio made by artificial intelligence to “clone” his voice…. Given Burnley’s dispute over the authenticity of the audio recording, the Court ordered supplemental briefing and conducted an evidentiary hearing on the issue. [emphasis added]

Mr. Walburn submitted a Declaration from Mr. Roskam affirming that the copy of the audio recording was an exact duplicate of the original.  Mr. Walburn also submitted a Declaration:

In it, Walburn declared that he has been “acquainted with [Burnley] for approximately 5 years[,]” and has “had the opportunity to engage in conversations with him and hear his voice on multiple occasions.” … He declared that he received a copy of the audio recording from Raskom on June 10, 2024, and that it is his “sworn opinion and belief that the audio recording is both authentic and that ‘Tyrone Jackson’s’ identity is [Burnley].” … He based his sworn opinion and belief on recognizing Burnley’s voice “from numerous prior verbal interactions with him.” … Finally, Walburn certified that “the recording in the Court’s possession is original or an exact duplicate of the original recorded evidence.”

Mr. Burnley responded:

In his response, Burnley maintains that: (1) the declarations of Walburn and Roskam are defective because referenced exhibits were not attached…; (2) that the declarations themselves must be stricken as submitted in bad faith pursuant to Civil Procedure Rule 56(h)…; (3) there are deficiencies in the chain of custody that require a certificate of authentication…; and (4) it would “be an abuse of judicial process to attempt to authenticate [the audio recording] on February 17, 2026 when the telephonic recording occurred on June 6, 2024…. Further, Burnley denies having any conversations with Walburn which would substantiate Walburn’s claim that he can recognize Burnley’s voice.

The court discussed the low burden of the authentication rule, Fed.R.Evid. 901(a) and applied that rule, writing:

The Court finds that Walburn sufficiently established the authenticity of the audio recording through the sworn declarations of Walburn and Roskam. The declarations primarily established: (1) how the recording was made; (2) how the recording was received; and (3) that the recording in the Court’s possession is an original or exact duplicate. The Court finds that none of these foundational facts depend on the attachment of an exhibit, as Burnley contends. Indeed, the absence of an exhibit, however referenced in the declaration, does not render sworn, factual statements inadmissible, nor does it undermine authentication under Rule 901. At any rate, the Walburn cured this deficiency by filing their Reply to Burnley’s Response and attaching the exhibits referenced in the Walburn and Roskam declarations. [emphasis added].

It also rejected Mr. Burnley’s “sham affidavit” argument. The affidavit did not conflict with any prior sworn testimony.

It then turned to Mr. Burnley’s chain of custody argument. He asserted “that federal law requires complete documentation of every individual who accessed the recording, including dates, times, and storage locations, and that failure to do so violates due process.”  Citing Fourth Circuit precedent, the court disagreed.  Chain of custody is not an “iron-clad requirement” and “the fact of a missing link” does not defeat authentication if there is “sufficient proof that the evidence is what it purports to be and has not been altered in any material respect.”  Mr. Roskam’s uncontroverted Declaration met that standard. It explained that the call was recorded in the ordinary course of business, transmitted internally, and provided to Mr. Walburn.  Both of the declarants testified that it was an exact duplicate.

In the court’s words, “[n]onetheless,” Mr. Burnley asserted that Mr. Roskam failed to explain “where the voice recording of June 6, 2024 was stored and the names of the LiveWire employees that had access to the recording” as well as file a “certificate of authentication.” The court held that Fed.R.Evid. 901 does not require certificate or formal attestations.

Burnley’s insistence on additional formalities finds no support in the Federal Rules or Fourth Circuit law. On the other hand, Rule 901(b) non-exhaustive list of evidence that satisfies Rule 901(a)’s authenticity requirement expressly states that “[t]estimony of a witness with knowledge” that “an item is what it is claimed to be” is sufficient. Fed. R. Evid. 901(b)(1). Here, the Court has been presented with two declarations, sworn under penalty of perjury, testifying that the audio recording is what it purports to be, which plainly satisfies Rule 901.

Holding that the “ultimate question is whether the authentication testimony was sufficiently complete so as to convince the court that it is improbable that the original item had been exchanged with another or otherwise tampered with,” the court found that the two sworn statements provided that foundation.

It also rejected the argument that it was “unfair and unjust” to permit authentication of a June 6, 2024, recording on February 17, 2026.  “Based on the evidence presented, which included two sworn declarations, the Court is satisfied that the audio recording is authentic and has not been materially altered.”

Finally, the court addressed the voice identification issue.  Mr. Walburn attested that he has known Mr. Burnley for five years, heard his voice many times, and recognized it on the recording. Mr. Burnley denied making the call, was skeptical of the voice recognition, and said that “under no circumstances [had he] had numerous conversations with [Walburn]…” However, he admitted that Mr. Walburn spoke to him in the past. The court wrote:

Given the issues raised by Burnley as to the authenticity and identity of the caller on the audio recording, the Court convened an evidentiary hearing on March 4, 2026. During the hearing, the Court heard the audio recording live…. Walburn testified that the recording was an accurate, unaltered copy of the recording he heard when his supervisor, Mr. Roskam, shared it with him…. Walburn clarified that he had previously worked for FedEx, was not fired for selling illegal narcotics, and does not know of anyone, including those who live on his street, named Tyrone Jackson…. He testified that he recognized the voice on the audio recording as Burnley’s voice because “it sounds exactly like him,” and that “[g]iven the previous lawsuits, his previous interactions calling FedEx, and all of the novel information in there, [there] is only one person, which is him.” … Walburn noted that he could recognize Burnley’s voice because Walburn had “two or three conversations with him” about “four or five years ago” and has heard him “sometimes outside or so.” …

On cross-examination, Burnley questioned Walburn about how Walburn recognized the caller as Burnley, to which Walburn responded: “I have ears” and “I can hear.” … The Court provided Burnley the opportunity to testify, under oath, that the voice on the recording was not his.… Notably, Burnley declined to testify under oath that it was not his voice on the recording….

Based on the evidence presented, the Court credits the statements by “Tyrone Jackson” on the audio recording as having been made by Burnley. First, despite being given the opportunity to testify in open court, under penalty of perjury, Burnley declined to state on the record that he did not call LiveWire on June 6, 2024 and that the statements captured on the audio recording were made by him. Second, Walburn sufficiently identified Burnley’s voice as the voice of “Tyrone Jackson” based on hearing Burnley’s voice on prior occasions as his neighbor. Third, Walburn notes that Burnley previously swore, in an affidavit, that Burnley had contacted Walburn’s former employer, FedEx, in 2023 and alleged that Walburn used illegal drugs and tried to sell Burnley illegal drugs “on numerous occasions.” … The Court considers this previous behavior as relevant to proving Burnley’s identity as the caller who lodged nearly identical accusations against Walburn in the phone call to Walburn’s current employer, LiveWire. See Fed. R. Evid. 404(b)(2) (permitting the use of prior acts for the admissible purpose of proving identity and not for proving propensity in conformance with the prior acts). [emphasis added].

Turning to the merits:

Here, the Court finds that Burnley’s use of a pseudonym demonstrates awareness of the contractual prohibition and an attempt to evade it. Such conduct indicates that Burnley not only circumvented the settlement agreement in bad faith, but willfully breached it. His unwillingness to testify under oath to the contrary further shows that Burnley is aware of the consequences of his actions but nonetheless refuses to take accountability for them. Given the seriousness of Burnley’s behavior in flagrantly violating the settlement agreement, the Court finds that enforcement is necessary.

The court addressed a number of other issues.

This blog was initially posted on  Electronic Discovery Reference Model and  JD Supra.

Plaintiff, X.AI, produces and develops A.I. models that it shares with the public. It filed suit to enjoin enforcement and moved for a preliminary injunction.  The motion was denied.  However, the court left the door open and wrote that the preliminary injunction decision was only a “threshold inquiry.”

THE SCOPE OF A.I. TRAINING IS IMPORTANT

The X.AI court’s analysis is of interest to general civil and criminal litigation because the scope of a GenAI training set may be important to laying, or challenging, an evidentiary foundation to a proffer of A.I. generated evidence at trial or on motions.  For example, if a training set is biased, the output may be biased.  See A Review of Sedona’s “Artificial Intelligence (AI) and the Practice of Law” by The Hon. Xavier Rodriguez (Sep. 27, 2023)(Sedona suggests: “AI evidence may require that the offering party disclose any training data used by the AI platform to generate the exhibit.”)(emphasis added).

THE CALIFORNIA TRAINING DATA TRANSPARENCY ACT

The X.AI court described the California statute:

The documentation must include “[a] high-level summary of the datasets used in the development of the generative artificial intelligence system or service” addressing, but not limited to, twelve enumerated topics…. Those topics include:

(1) The sources or owners of the datasets.

(2) A description of how the datasets further the intended purpose of the artificial intelligence system or service.

(3) The number of data points included in the datasets, which may be in general ranges, and with estimated figures for dynamic datasets.

(4) A description of the types of data points within the datasets….

(5) Whether the datasets include any data protected by copyright, trademark, or patent, or whether the datasets are entirely in the public domain.

(6) Whether the datasets were purchased or licensed by the developer.

(7) Whether the datasets include personal information ….

(8) Whether the datasets include aggregate consumer information ….

(9) Whether there was any cleaning, processing, or other modification to the datasets by the developer, including the intended purpose of those efforts in relation to the artificial intelligence system or service.

(10) The time period during which the data in the datasets were collected, including a notice if the data collection is ongoing.

(11) The dates the datasets were first used during the development of the artificial intelligence system or service.

(12) Whether the generative artificial intelligence system or service used or continuously uses synthetic data generation in its development….

Id. at *1-2.  However:

The statute exempts three types of models from disclosures: (1) a generative artificial intelligence system or service whose sole purpose is to help ensure security and integrity; (2) a generative artificial intelligence system or service whose sole purpose is the operation of aircraft in the national airspace; and (3) a generative artificial intelligence system or service developed for national security, military, or defense purposes that is made available only to a federal entity.

Id. at *2.

X.AI’s CLAIMS

X.AI is an entity subject to the statute.  It presented three arguments against the statute: “(1) that it violates the Takings Clause of the Fifth Amendment; (2) that it violates the First Amendment; and (3) that it is unconstitutionally vague.”

Presumably as a precaution, X.AI published a “high-level, limited disclosure that does not reveal its trade secrets.”  Id. at *2.  In is Complaint, however, it alleged a concern that the State Attorney General would assert non-compliance and seek to enforce the law.

Plaintiff alleges that several aspects of the datasets used to train AI models—including their contents, origins, size, and cleaning methods—are valuable and non-public…. Plaintiff alleges that “information about the datasets and processes AI developers use to train their AI models is a closely protected trade secret.”

Id. at *2. The X.AI court described and applied the legal standard governing motions for preliminary injunction.

THE COURT’S DECISIONS

The Attorney General of California was the defendant. I will, for ease of reference, even if not with fidelity to U.S. Const., Amend. 11, refer to the defendant as the “State.”

X.AI Has “Standing” to Challenge the Statute

The State contended that X.AI lacked Art. III “standing” to make the claim.  Oversimplifying, the State argued that X.AI had “engaged in some degree of compliance” and lacked injury in fact. The court rejected that defense. Id. at *3.

The Statute Did Not Violate the Takings Clause Because the Complaint was Generalized

The X.AI court next analyzed the “Takings Clause” argument.

Before the Court can evaluate whether Plaintiff has likely allegedly a successful Takings Clause claim, the Court must determine the likelihood of Plaintiff proving that the sources, sizes, and cleaning methods of its datasets qualify as trade secrets…. Under California law, “the test for a trade secret is whether the matter sought to be protected is information (1) that is valuable because it is unknown to others and (2) that the owner has attempted to keep secret.” [emphasis added].

However, the X.AI court found that X.AI’s supporting allegations were simply general statements. It wrote: “ Plaintiff’s Complaint trades in frequent abstraction and hypotheticals, rather than pleading specifics about Plaintiff’s practices.”  Id. at *4.  The court explained that:

When it comes to specificity, Plaintiff alleges that “[a]s part of its development process, xAI generally used the methodology outlined above.” … It offers that “xAI’s engineers invested substantial amounts of time and energy in acquiring datasets from various sources across the Internet to develop and eventually train the AI models that it has produced.” … Plaintiff also alleges that “[t]he amount of data that xAI uses is also valuable precisely because it is unknown to others” and that “xAI’s processes for cleaning, modifying, and refining the datasets it has obtained are economically valuable information too.”

Id.

Further, X.AI acknowledged that there is data overlap among many AI companies.  Instead, it asserted that the differences give a competitive edge.  The court wrote: “The problem is that Plaintiff has not alleged that it actually uses datasets that are unique, that it has meaningfully larger or smaller datasets than competitors, or that it cleans its datasets in unique ways. Plaintiff’s resort to generalizations and hypotheticals about the AI model development industry make it difficult for the Court to find that Plaintiff has carried the heavy burden of showing a likelihood of success in proving that trade secrets are at play here.”  Id. at *4.

The X.AI court acknowledged that, hypothetically, datasets could be trade secrets; however, here, X.AI’s allegations were only an “abstract pleading….” Id. at *5.  It held that: “Plaintiff has failed at this stage to sufficiently allege that trade secrets are implicated. As such, the Court finds that, as a threshold matter, Plaintiff is not likely to succeed on the merits of its Takings Clause claim based on the Complaint as pled.”

The Statute Did Not Violate the First Amendment

X.AI asserted that the statute compelled speech based on content and viewpoint.  “Specifically, Plaintiff alleges that [the California statute] is content-based because it requires Plaintiff to disclose specific content about its AI models, and it is viewpoint-based because it exempts developers of AI models related to network security, aircraft operations, and national security from its requirements.”  Id. at *5.

The X.AI court found that the statute is a “content-based speech regulation….”  Id. at *6.  However, it also found it to be “commercial speech.” Id. at *8.  It wrote that: “No part of the statute indicates any plan to regulate or censor models based on the datasets with which they are developed and trained.”  Id. at *7.

After a lengthy analysis, the X.AI court concluded:

Ultimately, Plaintiff has demonstrated a distinct possibility of prevailing on the merits…. But it had not demonstrated a likelihood of success on the merits. The information before the Court is insufficient to come to such a conclusion at this stage. Plaintiff therefore does not satisfy this threshold inquiry for a preliminary injunction on its First Amendment claim.

Id. at *8 (emphasis in original).

The California Statute is Not Void for Vagueness

The X.AI court wrote that the mandate of publishing “a high level summary: of the training datasets is not a picture of clarity standing alone….” Id. at *9.  However, that mandate is followed “by a precise list of the information to be included.” Id.

X.AI challenged terms such as “dataset” and “data point” as vague. It also argued that the list was non-exhaustive and therefore vague. The court disagreed.

Here, there is a list of information required akin to a set of factors—it is simply non-exhaustive. Given that a statute entirely lacking a list of factors can still be sufficiently clear, it is likely that a non-exhaustive list is enough.

Id. at *9.

The court viewed other vagueness challenges as “similarly insufficiently persuasive at this stage, absent a better-developed record, to find a likelihood of success on the merits…. Ultimately, the record at this stage is insufficiently developed for the Court to determine that Plaintiff is likely to succeed on the merits of its vagueness challenge.”  Id.

However, the X.AI court left the door open: “Evidence may arise during the course of litigation that eventually requires a different determination. But the pleadings and record as they stand are not enough at this time.”  Id.

COMMENT

Knowledge of how an A.I. tool was trained may be important in offering, or challenging, evidence at trial.  Admissibility may turn, at least in part, on knowing the data on which the AI was trained.  See M. Grossman & Hon. P. Grimm, “Judicial Approaches to Acknowledged and Unacknowledged AI-Generated Evidence,” 26 Col. Sci. & Tech. L. Rev. 110, 152 (2025).

ABA Formal Opinion 512 (“Generative Artificial Intelligence Tools” 2024) states: “The large language models underlying GAI tools use complex algorithms to create fluent text, yet GAI tools are only as good as their data and related infrastructure. If the quality, breadth, and sources of the underlying data on which a GAI tool is trained are limited or outdated or reflect biased content, the tool might produce unreliable, incomplete, or discriminatory results.” [emphasis added].

GenAI responds to prompts “based on patterns and structures learned from the data used to train the AI model.” Maryland State Bar Ass’n., “An Overview of Ethical Considerations for Attorney Use of Generative Artificial Intelligence Technologies” (undated), 3 (emphasis added).

The “quality of their training” may impact the limitations and risks presented by GenAI tools.  The Sedona Conference, The Sedona Canada Primer on Artificial Intelligence and the Practice of Law, 26 SEDONA CONF. J. 103,128 (forthcoming 2025), 128. A comprehensive and representative dataset is needed to train AI systems. Id. at146-47.  Inadequate training may lead to bias. Id. at 149-50.

Disclosure of training data may be an important predicate to admissibility. American Assoc. for the Advancement of Science, “Artificial Intelligence and the Courts” (2022), 12.

Facial recognition technology is a form of artificial intelligence.  In criminal cases, Maryland’s facial recognition technology statute requires a “description and the names of the databases searched….”  Md. Code Ann., Crim. Proc. Art. §2-504.

The ABA Task Force on Law and Artificial Intelligence Year 2 Report (2025) states: “Similarly, in healthcare, AI-trained devices may produce biased results if training datasets lack diversity, potentially leading to misdiagnoses.”

POSTSCRIPT

In Maryland, House Bill 823 “died” in committee.  It was titled “Generative Artificial Intelligence – Training Data Transparency.”  The official synopsis was: “Requiring a developer of a generative artificial intelligence system, on or before January 1, 2026, and before the developer releases or substantially modifies a certain generative artificial intelligence system, to publish on the developer’s website documentation detailing the data used to train the generative artificial intelligence system.”  For more information on legislative efforts, please see Bill to Create A.I. Evidence Clinic Pilot Program Was Vetoed in MD (Sep. 15, 2025).

The court wrote: “Legault’s motion to compel Costco to provide him with native files and metadata is denied because Costco states the files are security sensitive configuration files. While Legault argues that this is insufficient, Costco has offered to make the files available to Legault…. Additionally, the parties were unable to come to an agreement on a protective order, therefore, Legault’s argument about recourse of confidentiality concerns being through a protective order is denied.”

That’s all there is on this issue in the court’s opinion.

Bloomberglaw reports that the claim is an employment action for retaliation leading to termination. In ¶20 of the Complaint, plaintiff alleged irregularities in his personal devices “that indicated potential IT security issues.”  Plaintiff alleged that they “originated from Costco’s systems….”

Alleging no satisfactory response from Costco after presenting his concerns, plaintiff “downloaded files to his device so he could research these issues at home and report them to Costco again.” Compl., ¶34.  The files were configuration files. Id., ¶37.  He was suspended the next day and contacted by employees “who ran Cyber Operations for Costco.” Id., ¶35.

Bloomberglaw posted a copy of Costco’s response to the motion to compel. ECF No. 36.  It states:  “Costco terminated Legault’s employment after it discovered that Legault had removed, or ‘exfiltrated,’ approximately 6,000 files from Costco’s secure IT network.”

Costco wrote that, in the motion to compel, plaintiff wanted to review “configuration files” in native form.  It wrote: “Legault has demanded that Costco produce the ‘native files’ that he exfiltrated. Costco has produced over 4,000 of those files.”

Costco then asserted:

The remaining files, however, are security-sensitive “configuration” files. Configuration files are customizable instructions for software systems—they instruct the systems to do whatever they are designed to do based on the preferences and settings contained in the files. In his deposition, Legault acknowledged that a configuration file “provides a roadmap” of whatever software system it configures….  As stated in Costco’s investigation report regarding Legault’s exfiltration: “Configuration files could be used by an adversary to harm Costco’s network or Warehouse Operations.”

Costco wrote that it offered to make the files available for review at a deposition; however, it claims that plaintiff never responded and asserted that plaintiff “never raised the subject of reviewing the files during Helm’s deposition.”  Plaintiff disputed this; Costco asserts that it renewed the offer; and, Costco wrote that plaintiff never responded.

This blog was initially posted on  Electronic Discovery Reference Model.

The article states: “A settlement has been reached in a lawsuit that the former principal of Pikesville High School filed against Baltimore County Public Schools and the man convicted of using artificial intelligence to impersonate him and destroy his reputation.”

The terms of the settlement were not disclosed. The plaintiff “alleged that the school system refused to correct the record after an employee used AI to create an audio clip that sounded like him making racist and antisemitic comments.”  Id.

The fake audio clip contained racist and antisemitic remarks.  It was posted on social media.  Kristen Griffith, Faked by AI? Baltimore County Public Schools investigates offensive recording – The Baltimore Banner (Jan. 17, 2024). Mr. Eiswert was immediately removed from his  position as principal.

A criminal investigation followed, and:

Eiswert reported that he had conversations with Darien about not renewing his contract due to poor performance, inability to follow procedures and unwillingness to adhere to the chain of command.

Police tied Darien to a Gmail account that was used to send out the recording to two teachers with the subject line, “Pikesville Principal — Disturbing Recording.” The clip appeared on social media about 30 minutes later.

Forensic experts proved that the recording was fake.  Dylan Segelbaum and Kristen Griffith, Ex-Pikesville principal speaks after Dazhon Darien’s plea in AI case – The Baltimore Banner (Apr. 28, 2025).

Catalin Grigoras, an associate professor at the University of Colorado Denver and director of the National Center for Media Forensics, performed an “extensive analysis” and concluded that the audio contained traces of AI-generated content.

Grigoras also determined that someone then edited the clip and added background noises to make it seem more realistic.

Id.

Although the offensive audio was fake, the ex-principal testified that his life would never be the same. Dylan Segelbaum and Kristen Griffith, Ex-Pikesville principal speaks after Dazhon Darien’s plea in AI case – The Baltimore Banner (Apr. 28, 2025).  “The recording quickly spread across social media, and Eric Eiswert said his life forever changed.”  Id.

One day, Eiswert said, he was a respected educator who had served students and their families in Baltimore County for 27 years. The next day, people viewed him as a “racist and antisemite who said vile things about students and teachers.”

Though people later learned the recording was fake, Eiswert said, he knew “my life would never be the same.”

“I am slowly rebuilding my professional career, and I’m beginning to heal,” Eiswert said on Monday in the Baltimore County Courts Building. “But this experience will never leave me.”

Id.  The Banner reports that:

Eiswert received threats, including one that stated the “world would be a better place if you were on the other side of dirt.” He’s now principal of Sparrows Point Middle School in Edgemere.

Id.

The defendant entered an Alford plea to one count of disturbing school operations and was sentenced to four months in the Baltimore County Detention Center.  The sentencing judge said that a “good section of the community” will always look at Eiswert in a different light “whether it’s deserved or not.”  Id.  The judge added: “It is clear that he did not make that recording,” …. “He will always have to take that with him.” Id.

UPDATE: Racquel Bazos, “Former Pikesville high School principal settles suit with AI impersonator,” Baltimore Sun (Nov. 2, 2025)(“Darien was sentenced to four months in the county detention center for disturbing school operations.”).

A local business contacted police and provided an audio recording of the shooting.  Gunshots could be heard and a person shouted “Sconey, are you serious? You shot her Sconey.”

As a result, the police tried to find out who “Sconey” was.  Detective Jones performed an internet search, located two profile pages, and printed two “Instagram profile pages for ‘Scony_g’ and ‘Scony_da_g,’ each with a profile picture that resembles Spencer.”[1]

At trial, the State tried to prove that “Sconey” was the defendant, Mr. Spencer.  In addition to cell phone location data and DNA evidence, the State introduced the Instagram pages.

In response to the challenge to authentication, the State contended that the Instagram printouts were offered for a limited purpose and therefore authenticated.

In footnote 3, the Court wrote that “authentication of the mere existence of a social media profile is a relatively novel issue….”  [emphasis added].  It wrote that “our courts have not opined on it yet….”

As such, one may ask why the Spencer opinion, which decided a “relatively novel issue,” was unreported.  Maryland’s New and Improved Unreported Opinion Rule Does Not Go Far Enough – E-Discovery LLC (citing unreported decisions that were found “cert-worthy” by Maryland Supreme Court); Maryland Appellate Decision Permitting Limited Forensic Search of Non-Parties’ Cell Phones – E-Discovery LLC (discussing unreported decision on novel issue); Admissibility of Internet Searches About Terminating Pregnancy – E-Discovery LLC (same).

In Spencer:  “The State made clear that it was not offering the printouts to prove that Spencer had created the Instagram profiles, but only to show that the police had found the profiles while researching ‘Sconey.’”

The Appellate Court wrote: “Although our courts have not opined on it yet, it seems obvious that, … if the proponent merely wants to establish the existence of the social media evidence, they only need proof to support a finding that the social media evidence is what the proponent claims it to be.”  It stated:

There was sufficient proof from which a reasonable juror could find that the printouts are what the State purported them to be. The State proffered the printouts to show that the police had found the profiles while investigating the nickname “Sconey.” Because the State did not offer the printouts to show who had authored them, it was not necessary for the State to authenticate the source of the profiles. [emphasis added].

In short, because the profiles were offered for a limited purpose, they were properly authenticated.[2]

Mr. Spencer also argued that the pages could have been fabricated or “the photographs associated with the Instagram profiles could have been manipulated.”  The Court rejected that argument.  It wrote that the proponent need not exclude every possibility in order to authenticate evidence and this objection went only to weight. It wrote: “Because the State need not guarantee that the printouts are what it purports them to be, Spencer’s arguments are unpersuasive.” [emphasis in original].

As a precaution, the Court addressed harmless error, writing:

In this case, the printouts of the Instagram profile pages are cumulative of other properly admitted evidence. The printouts were admitted to prove that Spencer is “Sconey.” There was other evidence at trial, however, that makes this same point. First, the State introduced Detective Jones’s testimony in which he described YouTube videos that he found that tie Spencer to “Sconey.” In researching what the nickname “Sconey” was, Detective Jones testified that he “went onto YouTube … and came across music videos for Sconey G., otherwise known as Deamonte Spencer.” Detective Jones identified Spencer as the individual in the videos. Second, the State introduced significant evidence linking the nickname “Sconey” to Spencer’s phone. This evidence included: that in Kamira and Leah’s phones, Spencer’s phone number was listed as “Sconey”; that the billing subscriber listed for Spencer’s phone number was “Sconey Smith”; that Spencer’s phone sent Leah’s phone two audio files titled “Sconey G, Snakes in the Grass” and “Sconey G, blood in the streets.” Because this body of evidence repeatedly links “Sconey” to Spencer, it proves the same point as the printouts of the Instagram profile pages and is thus cumulative. Accordingly, the admission of the printouts, if erroneous, was harmless beyond a reasonable doubt.

Maryland has an extensive body of case law governing authentication of social media and video evidence:

• “Unfriending” Evidence – Maryland Follows “Reasonable Juror” Standard in Authentication of Social Media – E-Discovery LLC
• Authenticity and the Role of the Trial Court as Gatekeeper Under Md. Rule 5-104 – E-Discovery LLC
• Authentication of Asbestos-Containing Chalk Purchased on eBay – Exemplar Evidence – E-Discovery LLC
• Authentication of Surveillance Video by Lay Witness Under Silent Witness Doctrine – E-Discovery LLC
• Can a witness authenticate a video if the video contains images that the witness did not see? – E-Discovery LLC
• Authentication of Entire Video When Witness Observed Only Part of the Events Portrayed in the Video – E-Discovery LLC
• Expert Testimony Not Required to Show Ability to Disable Location Tracking on Smartphone – E-Discovery LLC
• Is a “Composite” Video Admissible and Can a Police Officer “Narrate” It at Trial? – Part 2 of 2 – E-Discovery LLC
• Surveillance Video of Robbery Was Properly Authenticated by Eyewitness – E-Discovery LLC

____

[1] The Court wrote that sometimes “Sconey” was spelled without an “e”.

[2] While I do not suggest that there was a need to discuss Griffin, the State’s approach in Spencer distinguishes it from Griffin v. State, 419 Md. 343 (2011).

WHAT IS EMAIL THREADING?

The court began with a description of email threading:

“Email threading is the technical process of recombining emails that comprise an email discussion, including replies and forwards.” The Sedona Conference, Commentary on Privilege Logs, 25 Sedona Conf. J. 221, 266 (2024). Threading, or grouping all emails together, may aid reviewers in identifying gaps in production or inconsistencies more efficiently. See, e.g., Pamela Garman & Casey Terrell, Technology and Ethics: Tips for the Tech-Averse in Effectively Managing ESI in Federal Court, 24 TYL 10, 11 (Spring 2020). The process identifies inclusive emails, meaning those “containing content that is not present in its entirety in any other email in the set of emails being analyzed.” The Sedona Conference, Commentary on Privilege Logs, supra at 266 n.72. Inclusive emails are typically “the last-in-time email in any branch of the thread, as well as any email with an attachment that is not also attached to a later-in-time email that contains the full content of the earlier email.” The Sedona Conference, The Sedona Conference Glossary: eDiscovery and Digital Information Management, 5th Ed., 21 Sedona Conf. J. 263, 381 (2020).

Id. at *2.  It added:

The remaining emails are categorized as non-inclusive because their content is included elsewhere in an inclusive thread. Id. at 343. Non-inclusive emails are often suppressed to increase the speed and efficiency of review and production of large numbers of emails. See Steven M. Medlock et al., Email Threading: Streamlining Discovery Review, 41 Comput. & Internet Lawyer 1, 1 (July-Aug. 2024). When non-inclusive emails are suppressed, the content is available in the inclusive email, but the metadata attached to each suppressed email is not included in production. See Joe Leonard, Threading the Labyrinth of Modern Email Collections, 46 DSBA Bar J. 12, 13 (March 2023).

Id. at *3 (emphasis added).

What may be lost in that process? The court wrote:

The Sedona Conference notes that the discovery duties of production for metadata “are generally the same” as for other ESI, meaning that “[m]etadata … must be produced when requested and not objected to.” The Sedona Conference, Commentary on Ethics & Metadata, 14 Sedona Conf. J. 169, 184 (2013) (citation omitted). Email metadata encompasses over 1,200 properties, including when the email was sent, or when a recipient received, viewed, replied, or forwarded the message. The Sedona Principles, 3d Ed., 19 Sedona Conf. J. 1, 210 (2018). Parties are encouraged to agree early in the discovery process on whether email threading should be allowed for review or production. Id. at 71-72 (Principle 3).

Id. at *3 (emphasis added).

NO BINDING AUTHORITY

Having laid the factual groundwork, the Tecfidera Antitrust court wrote that “although both parties argue that case law supports their positions, from this court’s perspective there is no binding precedent in the Seventh Circuit regarding this issue.”  Id. at *3.  The court reviewed the non-binding decisions in detail. Id. at *3-4.

It was, however, agreed by all that “producing threaded emails may be permitted (or prohibited) in a joint, stipulated ESI protocol.”  Id. at *3; see Fed.R.Civ.P. 29.

USABILITY

First, the court looked at “usability” writing: “The primary benefit to email threading is that it reduces the volume of emails the producing party must review as the process involves the de-duplication of non-unique email correspondence.”  Id. at *5.

However, Plaintiffs argued that the loss of metadata impeded usability.  In response, Defendant offered to write a script to generate a new field populated by that metadata.  Plaintiffs argued methodology that would impede data visualization. The court ruled as follows:

Despite Defendant’s assurances, the court sides with Plaintiffs. First, Plaintiffs have requested production of the emails “as they are ordinarily maintained,” which is the clearest default form of production set forth in Rule 34. Fed. R. Civ. P. 34(b)(2)(E)(ii). But Defendant proposes producing ESI in a form that deviates from the Rule 34 norm, from Plaintiffs’ request, and without demonstrating that its proposed alternative is “reasonably usable” as required. See id. This is insufficient to justify setting aside Plaintiffs’ proposal, which indisputably complies with the Rule.

Id. at *5.  In response to Defendant’s proposal to create a new metadata field, the Tecfidera Antitrust court wrote: “The fact that the metadata Plaintiffs seek can be provided in an additional field does not make it equivalent in terms of searchability and usability.”  Id.  It wrote that “Rule 34(b) does not state that a party can refuse to produce ESI as ordinarily maintained merely because the producing party’s alternative form is also used in discovery, see Fed. R. Civ. P. 34(b)(2)(E)(ii).”  Id. at *5.

The Tecfidera Antitrust court wrote that: “During the May 2025 motion hearing, Defendant argued that Plaintiffs have not provided details to substantiate their claim that email threading would obstruct searchability, but this assertion incorrectly places the burden on Plaintiffs….  Rule 34(b) does not require that a party seeking relevant discovery justify its request that ESI be produced in the form in which it is ordinarily maintained.”  Id. at *6.

PROPORTIONALITY

Second, the court analyzed proportionality.  Defendant argued, supported by an expert affidavit, that threading strikes a balance between reducing its costs, while providing “all substantive information…..” Id. at *6.

Plaintiffs responded that the defense “has not substantiated its claims of substantial burden with any details regarding the amount of time required for review or any specifics regarding potential additional hosting costs.”  Id. at *6.  The court wrote:

The court agrees with Plaintiffs on this point. Defendant provides no analysis of the proportionality factors in Rule 26(b)(1) beyond unspecified claims of substantial burden that the potential scope of discovery will exacerbate…. [The Defendant’s expert’s] testimony that threading “can materially reduce the number of documents for review … result[ing] in lower document hosting costs and quicker completion” lacks sufficient specificity to assess Defendant’s alleged burden…. Likewise, that the lack of threading “can result in higher hosting costs” is devoid of both specificity and certainty…. Nonspecific or speculative claims are insufficient for the court to weigh Defendant’s burden against the remaining proportionality factors in Rule 26(b)(1).

Id. at *6.  In my words, the Plaintiffs said that the defense lacked sufficient metrics and the court agreed.

THE DOOR WAS LEFT OPEN

However, the court left the door open for “more evidence of a specific substantial burden….” Id. at *7.  It added:

And a more detailed argument regarding “the importance of the [metadata] in resolving the issues” given the existence of other tools to increase searchability might change the calculus of a proportionality analysis. Fed. R. Civ. P. 26(b)(1). But as it stands now, Defendant fails to meet its burden to show that its objection on grounds of proportionality.

Id. at *7.  It denied Defendants’ threading motion without prejudice.

USE OF THREADED EMAILS IN DEPOSITION AND TRIAL

In Footnote 2, the Tecfidera Antitrust court addressed, but did not decide, a very interesting argument:

Plaintiffs also argue that excluding non-inclusive emails disadvantages trial strategy because they will be forced to present entire email threads to witnesses during depositions and examination at trial…. Because this court finds that Plaintiffs are entitled to all metadata available in the emails as requested and ordinarily maintained, this argument need not be addressed. That said, this court is skeptical of this argument. Defendant expressed willingness to agree to a protocol permitting the use of redacted emails for presentation to witnesses…. Plaintiffs cite no authority supporting the proposition that the rules of discovery require consideration of a preferred method of presenting exhibits to witnesses when determining the form of production, and they do not argue that the loss of metadata in non-inclusive emails would impact this task.

Id. at n. 2 (emphasis added).

I suggest that, if agreement on threading cannot be reached, Tecfidera Antitrust holds that a party seeking to employ threading over objection must provide specific metrics in support.

For what it is worth, I am more intrigued by the Footnote 2 argument.  I suggest that many forms of ESI – such as email chains, text message chains, chats, “modern attachments,” Excel workbooks, PDF portfolios, and many more – present an important question: What is a “Document?”

Also, I am not convinced that showing a redacted email chain to a jury is equivalent to showing a focused and unredacted one. While not precisely the same context, it has been said that:  “[R]edacting allegedly nonresponsive or irrelevant portions of discoverable documents ‘breed[s] suspicions.’” In re State Street Bank & Trust Co. Fixed Income Funds Inv. Litig., 2009 WL 1026013, at *1 (S.D.N.Y. Apr. 8, 2009), as quoted in Relevance Redactions Rejected – Rule 26(f) Resolution; accord Beverley v. New York City Health & Hosps. Corp., 2024 WL 1342806, at *10 (S.D.N.Y. Mar. 29, 2024); Castlelake, L.P. v. Lancashire Airline War Consortium/Airline Hull, 2024 WL 4007498, at *3 (Minn.Dist.Ct. Aug. 05, 2024); Conservation L. Found., Inc. v. Shell Oil Co., 2023 WL 5434760, at *6 (D. Conn. Aug. 22, 2023).

I think that the evidentiary argument in Footnote 2 has more force.  For more on “what is a document?,” please see:

• What is a “Document?”: Interior Email Omitted from Email Chain – Sanctions Follow for Lack of Candor to Court;
• What is a Document? (Part II);
• What is a Document? (Part III);
• ESI Protocol Should Define “Documents” and Address Redaction Based on Irrelevancy; Sedona Conference Commentary on Discovery of Collaboration Platforms – What is a Document?;
• “Modern Attachments” or “Pointers”- What is a Document? (Part IV); Uber Technologies – Another Hyperlink Decision.

This blog was initially posted on  Electronic Discovery Reference Model.

There is a substantial and growing body of scholarly and regulatory writing applicable to the use of digitally-obtained geographical information in other contexts, including civil litigation, information governance, and regulation.  It would take multiple books to reach even the tip of the scholarly iceberg.  This blog does not seek to climb that mountain.  Instead, like a first-year law school class, it is intended as issue-flagging.

GEOFENCING IS BEGINNING TO APPEAR IN CIVIL LITIGATION

Geofencing is beginning to appear in civil litigation.  In KalshiEX, LLC v. Hendrick, 2025 WL 1073495, at *1 (D. Nev. Apr. 9, 2025), a plaintiff moved to enjoin State agencies “from pursuing civil or criminal enforcement against Kalshi for offering event contracts in Nevada.”  The State had sent Kalshi a cease-and-desist letter alleging that Kalshi’s actions were unlawful in Nevada.

Injunctive relief was granted for a number of reasons; however, in pertinent part: “Kalshi presents credible evidence that even if it could implement geofencing at great expense, it could not do so immediately as the defendants demanded. And, as discussed at the injunction hearing, there are questions about whether Kalshi could recover monetary damages against the defendants in either state or federal court and, even if it could, whether those damages would be capped such that Kalshi could not meaningfully be compensated for the millions of dollars it asserts it would have to spend to geofence out Nevada market participants.” Id. at *7 (emphasis added).

USE OF GEOTARGETTING POTENTIAL JURORS

One of the most fascinating uses of geofencing in civil litigation was an attempt to reach potential jurors by targeting advertisements around the courthouse in one of several hundred “Roundup” cases. Pilliod v. Vionsanto Co., 2019 Cal. Super. LEXIS 23599, at *7 (Alameda Cty. Super. Ct. Apr. 4, 2019)(unpublished).

Plaintiffs alleged personal injury caused by Monsanto’s Roundup product. The case was set for a jury trial “against the backdrop of substantial press coverage outside the courtroom of similar cases….”

Plaintiffs moved to preclude Monsanto from publishing advertisements about product-related issues.  Among the various types of advertising, “at the hearing on 4/4/19, counsel for plaintiff asserted that Monsanto was using geofencing or geomarketing to target advertisements about the safety of Roundup to cellphones and other mobile electronic devices that in the area of the courthouse. Plaintiff’s assert that  this targeted advertising on a central issue in this case is jury tampering.” Id. at *6-7.  Plaintiffs’ counsel asserted, without evidence, that “Monsanto is geomarketing to the courthouse.”

The Pilliod court wrote:

Assuming arguable misconduct, there are practical concerns. In contrast to persons in a courtroom with placards or buttons, the court cannot readily observe geomarketing. The court could, but will not, order Plaintiff’s and Monsanto to cease and desist from geomarketing the courthouse and to submit declarations of compliance. Even if issued, any such order precluding geomarketing would not preclude general marketing in Alameda County and the Bay Area, and the jurors might be exposed to such marketing at home and on their ways to and from the courthouse. The court is not persuaded that the alleged geomarketing is materially different from carrying signs outside a courthouse or carrying placards or wearing  buttons inside a courtroom or that it requires a different judicial response.

Id. at *7-8 (emphasis added). With the exception of Ms. Neff’s article which brought this decision to my attention, Shepard’s does not report any subsequent citation of this decision.

GEOFENCING, GEOTARGETING, AND GEO-BLOCKING

IN SOME CIVIL CONTEXTS

One recent article defines and discusses the role of geofencing, geotargeting, and geo-blocking in some civil contexts.  S. Neff, “Thinking Inside the Box: Geo-Fencing Technology Litigation and Lessons for the Consumer Financial Services Industry,” 78 Consumer Finance Law Quarterly Report 1 (2025).[1]

Oversimplified, geofencing occurs when law enforcement knows where and when a crime was committed and law enforcement serves a multi-step geofence search warrant to find out what cell phones were in the “geofenced” area at the relevant time, thus locating suspects.  Uses in civil litigation can be hypothesized.  X claims that A harassed X at 4 p.m., in the employer’s office building.  Can geolocation data show that A – or, at least, A’s cell phone – was, or was not, at that location at that time?[2]

Geotargeting identifies devices currently in an area so that targeted marketing can be sent to those devices. A shopper in Store Y may receive a text about a special sale item located in Aisle 9 of that store while the customer is still shopping.

Geo-blocking is used to prevent access.  For example, a seller that does not want to purposely avail itself of a particular forum may geo-block its website or advertising from that forum. And, “[i]n Ukraine, SpaceX used location data to create a geofence around Crimea so that the Ukrainian military could not access Starlink in that location because Musk did not want Starlink to be used offensively by Ukraine to gain territory that Russia controlled.”  Gavin Small, 20 Duke J. Const. L. & Pub. Pol’y Sidebar 107, 113 (2025).

Geofencing may be more widespread than most of us recognize.  In “Thinking Inside the Box,” Ms. Neff wrote:

Many financial service companies have adopted the use of geo-fencing and geo-location technologies. These products and services take advantage of these emerging technologies to improve the services provided and contour the products offered to consumers. While these technologies have much to offer, geo-fencing and geo-location are more frequently becoming the subject (or at least an aspect) of litigation. This Article explores the ways that geo-fencing and geo-location have found a presence in the American courtroom and the lessons this development has for consumer financial service companies. [emphasis added].

Some commercial uses of geofencing are:

For example, consider a customer who enters a grocery store and crosses into an invisible geo-fence. While in the store, he may receive a text message with a linked coupon or a message about weekly specials or may see other focused advertising on social media. This is a prime example of an advertising geo-fence in action–providing targeted information and opportunities to a consumer as he enters the designated geo-fence area. In addition to advertising, geo-fences serve a variety of purposes, such as gathering data on location behavior, recording employee attendance, and improving resource management based on how locations are frequented. Importantly, geo-fences are increasingly used to ensure compliance with regulations. For example, Geo-fences can verify the location of a person placing a wager on a state-regulated wagering platform, ensuring that the user is not located in a county or parish that has expressly prohibiting such gaming. Similarly, geo-fences are used to limit the territory of internet-streamed radio broadcasts, helping unlicensed stations stay within the 150-mile exemption of the U.S. Copyright Act for streaming broadcasts online. Geo-fences also play a role in security and safety. States have created grant programs to use geo-fencing to protect schools from security threats. Additionally, geo-fences are used to guide commercial drivers away from restricted areas during political events or to track commercial boats for purposes of reducing pollution.

Geo-targeting is a similar, but different concept: “While geo-fencing is triggered by a device moving into or out of a designated geo-fence, geotargeting focuses on the current location of a device, regardless of movement…. In the context of digital environments, such as internet browsers and online services, geotargeting can be used to tailor content and advertisements to a user’s location to enhance relevance and effectiveness.  Ms. Neff wrote:

For example, when a user in Chicago opens the browser on her cell phone and searches for “Chinese restaurant,” the browser geotargets the user’s location and assumes she is searching for Chinese restaurants in Chicago, which is reflected [in] the search results. Similarly, the Federal Communications Commission authorized the Wireless Emergency Alerts System to use geotargeting to send messages to all devices within a specified area to receive emergency alerts. This is how a user receives AMBER Alerts or emergency shelter alters targeted to the user’s current location.

Ms. Neff asks: “What lessons do criminal cases about geo-fencing warrants have for consumer financial services companies?”  For example, she points out that any company using geofencing as a marketing or other commercial tool may find itself served with a search warrant for the data.  That is an information governance consideration.  Does management wish to take that risk? [3]

Geo-blocking, Ms. Neff wrote, is “using location technology to prevent access to websites or information” and she opines that it and geofencing are “some of the latest battlegrounds in personal jurisdiction fights, especially in cases involving e-commerce or internet jurisdiction.” Personal jurisdiction may be demonstrated by action purposely aimed at a forum state, such as seeking to contract or provide goods in that forum.  Ms. Neff asks:

But what if a defendant uses geo-fencing or geo-blocking technology to intentionally avoid engaging with or doing business in the subject forum? The general sentiment is that the extent of a defendant’s efforts to utilize location technology to avoid transacting in a certain jurisdiction tend to demonstrate that a defendant did not purposefully avail itself of the privilege of conducting business in that for[u]m. Efforts to block contacts with a forum– even if not foolproof–demonstrate a defendant’s effort to avoid minimum contacts within a forum, and that the defendant has a due process interest in not being hailed into court in that forum. “To be sure, the proposition that a website’s affirmative geoblocking efforts should weigh against the exercise of personal jurisdiction is unobjectionable.” [emphasis added; citation omitted].

However, Ms. Neff asserts that “the reverse is not necessarily true,” and a failure to geoblock is not a purposeful availment of the forum.[4]

The Duke Journal article notes that several states bar geofencing around physical and mental health care facilities.  Geofencing can be used to “promote targeted products and services….”  As one example, the Neff article states that, absent a legal prohibition: “Digital advertising firms can set up geofences around health care entities, and once a person crosses that invisible barrier, the person is bombarded with text messages and advertisements, urging the person not to seek reproductive or gender-affirming care.” [citation omitted].  While some states have barred the process, Ms. Neff wrote that others have not.

After discussing Pilliod v. Vionsanto Co., 2019 Cal. Super. LEXIS 23599 (Alameda Cty. Super. Ct. Apr. 4, 2019)(unpublished), Ms. Neff concludes:

For consumer finance companies engaging in high stakes litigation, geofencing targeted messages about a company’s commitment to transparency and fair practices could be a powerful tool to augment the conversation in advance of a jury trial. That said, the more narrowly the geographic area is targeted, the less likely it will survive scrutiny.

As a matter of information governance, if nothing else, entities planning to use geofencing, geotargeting, or geo-blocking should make an informed analysis.

A SAMPLING OF GEOFENCING AND ITS COUSINS

IN CIVIL CONTEXTS

As noted at the outset, this topic could fill books and is the subject of numerous law review and journal articles.  A few samples follow:

****

Lawyer Advertising

“’Weird, this is the fourth ad I have gotten on my phone from a personal injury firm since I’ve been here. It’s like this thing can read my mind …,’ Tanner thinks as he browses his phone while he sits in the University Hospital waiting room. He is waiting to see his elderly father, Cole, who just had a fall at the supermarket.”  Trevor Warren, “En Garde!: Geofencing and the Pressing Need to Update ABA Model Rule 7.3,” 63 U. Louisville L. Rev. 187, 187 (2024).  “This technology is already being used by retail giants such as Target.”  Id.  The article states: “With the realization of a growing need for regulation of attorneys using geofencing to solicit their legal services, and in light of a silent ABA, some states have issued their own advisory opinions.” Id. at 204.

****

Privacy and Health Care

“In September 2017, the Attorney General of Massachusetts settled its suit against Massachusetts-based advertising company, Copley Advertising, LLC, for violating state consumer protection laws. Copley had created a ‘geofence’ around Massachusetts reproductive health facilities, tracking consumers’ physical location and disclosing that location to third-party advertisers.”  Lianne Foley, “Targeted Advertising in the Healthcare Industry: Predicted Privacy Concerns,” 27 Annals Health L. Advance Directive 62, 62 (2017).  “[Copley Advertising] can set up a mobile geofence around any area.”  Id. at 65 (emphasis added); see Hana Ferrero, “Identifiable to Whom? Clarifying Biometric Privacy Rights in Illinois and Beyond,” 92 U. Chi. L. Rev. 1027, 1040, n. 79 (2025)(“Prohibiting the use of geofences helps ensure patients’ privacy regarding visits to healthcare facilities (e.g., abortion clinics, substance abuse treatment, and mental health clinics) and their underlying medical conditions.”).

****

Reproductive Rights Privacy & Political Campaigns

“Targeted advertising is a critical, hotly contested policy issue for at least four reasons.”  Derek E. Bambauer, “Target(Ed) Advertising,” 58 U.C. Davis L. Rev. 1429, 1435, 1509 (2025):

“One Wisconsin group used geofencing techniques to identify the phones brought into the offices of health care providers such as Planned Parenthood and then used the device IDs to target ads to their owners on social media platforms such as Facebook and Snapchat. Similar campaigns ran in New Jersey, California, Florida, and Colorado. Abortion rights groups expressed concerns that just the use of targeted advertising in this context is problematic from a health privacy perspective. And while advertising can often lead to outcomes that are beneficial for all involved, such as when a consumer purchases a product that they did not previously know about but ultimately find useful, the abortion geofencing campaign is likely one that is a zero-sum game….  [R]esearchers used targeted advertising to display ads about the risks of climate change to members of the Republican Party in a pair of competitive Congressional districts….”

****

Abortion Clinics

“Protesters have used geofencing so that third party data brokers bombard abortion clinic visitors’ phones and devices with targeted information about crisis pregnancy centers and adoption agencies. Anti-abortion groups can also purchase geolocation datasets for a specific location, such as Planned Parenthood facilities, from data brokers. The information these brokers provide is anonymized, but does provide assumptions about where the smartphones that visited the chosen location spend the night specified by census tract.”  Carson Goos, Elaine McCabe, “Abortion Protesting,” 26 Geo. J. Gender & L. 245, 260 (2025)

****

Employment Issues

Another article wrote:

It’s 10:00 p.m. Do you know where your employees are? For an increasing number of employers, the answer is yes. The location data collected from employees’ smartphones makes it easy, and increasingly inexpensive, to track the whereabouts of employees whenever their phones are with them.

The average person has more than eighty apps downloaded on their smartphone and never leaves home without that phone. Of those eighty-plus apps, many are recording that phone’s location every few seconds. That information is then sold to third-party data collectors, who sell that location data in a lucrative and growing open market. This practice has turned into a $21 billion location data industry that is predicted to more than double to $53 billion by 2030.

Although this data is touted as being anonymized, with the right knowledge and motivation, it is easy to figure out which data sets belong to a specific person. In 2018, the New York Times tested this by trying to figure out which data set, out of millions, belonged to the smartphone of Lisa Magrin, a forty-six-year-old math teacher. It did not take long for them to discover that only one device traveled from Magrin’s home in upstate New York at 7:00 a.m. to her school fourteen miles away and then back to her home late that afternoon. That was sufficient for the Times to track all her smartphone’s movements. Ms. Magrin was disturbed to discover that the Times reporters had learned when she went to Weight Watchers, had a doctor’s visit for a medical procedure, took her dog on a hike, and spent a night at her ex-boyfriend’s house.

Elizabeth A. Brown & Matthew M. Cummings, “Every Step You Take: Securing Employees’ Location Data Privacy,” 26 Minn. J.L. Sci. & Tech. 1, 4–5 (2025).

“In July 2021, the Catholic news outlet Pillar outed a prominent Catholic priest using location data to discover that he attended LGBTQ+ bars and nightclubs, forcing him to resign.

In 2022, ByteDance, TikTok’s parent company, admitted to firing four employees after it learned those employees were using location data to track the whereabouts and relative proximity of journalists and coworkers they suspected of whistleblowing.”  Id. at 5.

“For example, if an employer is aware of a unionization drive, they could geofence that area and see who attended it.”  Id. at *6.

****

Employment Issues (Cont.)

“Employers may use geofencing to receive alerts when an employee, equipment, or electronic records enters or exits a defined area. Employers may protect assets, track employees’ location, track employees’ performance, e.g., deliveries per hour, and track hours worked. A geofence may also alert an employer to an unauthorized entry or unauthorized equipment removal and may exclude equipment. An employer can link geofences to equipment, so that it stops working when removed from that area. An employee may use a geofence to protect his privacy or property.”  William E. Hartsfield, 1 Investig. Employee Conduct § 6:37.

****

Regulation of Drones and Use at Major Events

“[S]cholars … propose that the adoption of new geofencing technology will solve the problems created by the introduction of widespread drone usage. In particular, this would obviate the need for registration, and for the FAA to place undue restrictions on issuing beyond visual line of sight (BVLOS) permits…^.  Geofencing may indeed be the future for drone manufacturers. Unlike the current regulatory regime, it solves a myriad of problems posed by drones which are not properly addressed under the current regulatory scheme.”  John Ricci, “No Drone Zones: Assessing the FAA’s Role in Implementing Geofencing and the Future of Drone Regulations in the United States,” U. Ill. J.L. Tech. & Pol’y, Spring 2022, at 161, 163.

“Developers can create geofences which can be read by any device containing an onboard GPS system with the requisite software to detect the geofence. Once triggered, the geofence can be used to serve any number of functions from sending text messages, triggering events in mobile games, or even targeting advertising.”  Id. at 168.

“ Geofencing is a far more effective solution to the problem of airspace safety than anything that the FAA has attempted to date….  For example, geofencing technology can help keep drones out of dangerous areas…. Temporary geofences can be set up around huge public events (e.g., the Super Bowl, parades, or concerts), hazardous locations (e.g., wildfires, pipeline explosions, nuclear plants), or people (e.g., the president).” Steve Calandrillo, et. al., “Deadly Drones? Why FAA Regulations Miss the Mark on Drone Safety,” 23 Stan. Tech. L. Rev. 182, 188, 221, 247 (2020).

“Geographical restrictions also found a particularly striking application in the drone industry. DJI, a leading drone manufacturer, employed a Geospatial Environment Online (GEO) system that was built into every one of its drones and which used GPS data to enforce no-fly zones near sensitive locations. DJI’s geofencing system categorized airspace into five zones by color: Red (Restricted, no-fly), Blue (Authorization, unlockable with proof), Orange (Enhanced Warning, extra confirmation required), Yellow (Warning, alerts but no blocks), and Gray (Altitude, where flight is allowed but limited to a specific maximum height).”  João Marinotti, “Defragging Ownership: How Corporations Sliced, Diced, and Sold the Bundle,” 100 Ind. L.J. 1931, 1951 (2025).

****

Copyright Royalties

Courts should allow broadcasters to use geofencing to avoid copyright royalties.  Bradley Ryba, “Iheartgeo-Fencing?: The Section 114 Exemption That Illustrates Why Full Sound Recording Rights Are the Sine Qua Non for A Vibrant Music Industry,” 20 Marq. Intell. Prop. L. Rev. 33, 39 (2016).

****

Franchise Regulation and Targeting

“Today’s franchise marketing may involve ‘traditional’ advertising, such as advertisement in franchise trade publications or in local or regional business publications; but more often, franchisors are utilizing the technologies of the twenty-first century, such as ‘Google Ads,’ ‘Facebook Ads,’ ‘re-marketing,’ or ‘geo-fencing’ or ‘geo-targeting.’ As applied to digital advertising, the current state of franchise advertising regulations is archaic and antiquated … and frankly inapplicable in most situations.”  Mark J. Burzych, “Franchise Advertising in the Digital Age: Regulators Need to Contemporaneously Address Advancing Advertising Technologies or Step Aside,” 40 Franchise L.J. 221 (2020).

“As described in further detail below, popular advertising technology such as Google Ads, retargeting, remarketing, and geo-targeting, or geo-fencing does not squarely fit in the current regulatory scheme.” Id.

Advertisers can maximize their return on investment (ROI) and click conversions on online ads through geo-targeting and geo-fencing. Geo-targeting is simply customizing an ad campaign to reach audiences (through their computers or mobile devices) who are physically located in a particular geographic area. Both Google Ads and Facebook Ads offer advertisers the option to limit a particular campaign to a geographic area, such as the United States or Michigan. If advertisers want to get even more geographically specific with their ad audiences, they can pursue a geo-fencing strategy.

Geo-fencing allows advertisers to create a virtual box around a geographic location, wherein their ads will appear only on computers or mobile devices that are physically located in the box. Using GPS or Internet users’ IP addresses, advertisers can reach out only to specific users who enter a geographic area relevant to the advertiser’s business. For example, a retail establishment in a shopping mall may draw a several-block radius around the mall to specifically target Internet users who may either be out in the vicinity of the mall, or are already shopping inside. Alternatively, a local pizzeria (say one of two in town) can draw a geo-fence around not only its own location, but also around the location of its competitor, specifically targeting people in the vicinity of its competitor with advertisements showing its comparatively lower prices or current promotional deals.

Many companies take advantage of geo-fencing when customers download their mobile app and opt-in to the app’s “location services.” By allowing GPS to track where the device is located, advertisers can send push notifications, text messages, or PPC (pay-per-click) ads to the device. However, geo-fencing is available to companies that do not have a mobile app to track customers’ movements, and any business with access to Google or Facebook can set up a geo-fencing marketing campaign.

Advertisers can set up geo-fencing campaigns through either Google or Facebook relatively easily. Google Ads users can manually enter a specific location to target their ad campaigns, rather than selecting an entire state or country. Alternatively, advertisers may choose to target a radius of miles around a specific location, allowing the advertisers to maximize the specificity of their targeted audience.

Facebook users have an identical option. When customizing their Facebook ad campaign, advertisers can target their ads to a specified location, such as a particular state or country. If the advertiser wants to be even more specific, Facebook provides an interactive map that allows the advertiser to drop a “pin” on a central location and determine a radius of miles surrounding the pin where the ad campaign will be targeted. Although the minimum radius an advertiser can draw is one mile, advertisers can be even more specific by excluding certain geographic areas captured by the radius.

Id. at 233-34.

****

Some Regulatory Efforts

“Though no federal standard yet exists for geolocation data tracking, twelve states have enacted statutes that address data privacy practices in regard to consumers’ geolocation data. Many consider the Virginia Consumer Data Protection Act (‘VCDPA’), which took effect on January 1, 2023, as one of the leading state approaches on the topic. The VCDPA requires controllers of data to obtain affirmative consent from Virginians to process and sell their sensitive data, including their precise geolocation data.”  Eliza Smith-Driggs, “The ‘Right to Be Let Alone’ Should Apply to Geolocation Tracking Within the Home,” 30 Rich. J.L. & Tech. 495, 4 (2024).

“This Note also recommends that Virginia legislators require companies to enable a geofence around a consumer’s home so the consumer can choose when she wants her data collected….  To encourage even more control and home privacy, smart devices could ask homeowners whether they want their geolocation data tracking services turned off once they enter their home…. The default should be that the geofence blocks all trackers. But, for example, if one of the trackers is the homeowner’s mobile carrier needing location data to send a cell signal, the homeowner should be able to manually allow her carrier into the geofence to receive cell service.”  Id. at 65.

****

“Buying a Way” Around the Fourth Amendment?

“Well before Dobbs leaked, the Secret Service, IRS, U.S. military, DEA, ICE, CBP, FBI, and governmental entities at the state and local levels had all purchased data from brokers without obtaining Fourth Amendment search warrants. Lawmakers had already begun investigating these entities for “buy[ing] [their] way around the Fourth Amendment.”  Andrew Wade, “The Clocks Are Striking Thirteen: Congress, Not Courts, Must Save Us from Government Surveillance Via Data Brokers,” 102 Tex. L. Rev. 1099, 1101–02 (2024)(cleaned up).

****

Geofence Data Must Be Correctly Interpreted to Avoid Error

Geofence data tracks a device, not a person.  “For Jorge Molina, his wrongful arrest arose when a geofence search revealed that a cell phone logged into his accounts was present at the time of a murder; police rushed to arrest him, ignoring evidence that his abusive stepfather used his phone and car.”  Mary D. Fan, “Suspecting with Data,” 109 Minn. L. Rev. 2253, 2256, 2262-63 (2025)(emphasis added).

“The arrest at Molina’s workplace shattered the young man’s life, resulting in his mugshot and name plastered over the news as a killer, six days of incarceration, and an arrest in his records that jeopardized his career and educational goals.”  Id.

****

Contract Enforcement

“Geofencing allows lenders and dealers to wield significant control over drivers post-transaction. Such features enable lenders to disable a car once it exits a predetermined area established by the lender. Once a lender receives an alert of the violation, the lender can remotely disable the vehicle to limit the driver’s mobility.”  Stacy-Ann Elvy, “The Vehicle Monitoring and Collection Technology Era,” 110 Iowa L. Rev. 43, 69 (2024).

This blog was initially posted on  Electronic Discovery Reference Model and  JD Supra.

_____

[1] This blog summarizes Ms. Neff’s article.  It is based on, and all quotations are from, that article, unless otherwise indicated.

[2] As used herein, this blog does not discuss obtaining user location data directly from a user’s device through either a search warrant or civil discovery.  It discusses obtaining location data that is collected and stored elsewhere.

[3] For discussion of geofence search warrants, please see Certiorari Petition to U.S. Supreme Court in 4th Circuit Geofence Decision – E-Discovery LLC; En Banc Fourth Circuit Geofence Decision is Splintered – E-Discovery LLC; New Jersey Decision Permits Geofence Warrants Upon a Proper Predicate – E-Discovery LLC; Texas Upholds Narrow Geofence Warrant After Melee Resulted in Murder – E-Discovery LLC; Possible Clarity About Geofence Search Warrants – E-Discovery LLC; Geofence Search Warrant Held Valid – E-Discovery LLC (Maryland); Keeping an Eye on Geofence Warrants – E-Discovery LLC; Requests for “Tower-Dump” or “Tower-Extraction” Search Warrants Declined Under Geofence Precedent – E-Discovery LLC.

[4] One authority cited by Ms. Neff states: “[E]quating ‘a failure to geoblock with purposeful availment’ would ‘effectively mandate geoblocking for any website operator wishing to avoid suit in the United States.’ …. Safex cites no authority suggesting that such ‘a failure to act might constitute purposeful availment,’….”  Safex Found., Inc. v. SafeLaunch Ventures Ltd., 694 F. Supp. 3d 1, 17 (D.D.C. 2023).

In October 2020, the Denver Police Department hit a wall. The Department was investigating an arson at a home of eight individuals, killing five who were trapped inside. Police executed at least twenty-three search warrants, four of which searched through thousands of people’s locations via their mobile devices. Yet still, police officers could not pinpoint their suspect. Police then decided to follow in the footsteps of other departments around the nation; they filed three warrants to uncover any Google searches that included the address of where the fire occurred.

Id. at 977-78 (emphasis added).

Keyword warrants compel search engines to hand over the personal data of users who have searched specific terms, within a certain timeframe, sometimes within a certain area.

Id. at 979 (emphasis added).

Ms. Edano notes: “Keyword warrants do not demand the search history of a specific person, but rather, keyword warrants demand the information of persons who have searched for a single or various keywords. These demands for information are not tied to a crime in terms of location or any other tangible piece of evidence. Instead, keyword warrants suspect individuals who shared a common interest in specific words or addresses.”  Id. at 984.

The article states that: “Google, the most popular search engine, retains data on users’ identities and what users search, making it the ideal target of these warrants.”  Id. at 979-80.

Google retains incredibly detailed personal data. Google collects personal identifiers (name, phone number, and address), demographic information, commercial information, biometric information (if provided), internet and network information (search terms, content and advertisement views, interaction with applications, browsers, devices, IP addresses, and activity on third-party sites that use Google services), geolocation data, voice and audio information (if provided), professional or educational information (if using an account maintained through an organization), photos and videos, and emails and documents. Google generally protects private information by only allowing the user to see this data and shielding user data from “unauthorized access, alteration, disclosure, or destruction of information” Google keeps. As of 2020, Google began automatically deleting users’ location history and web and application activity after eighteen months for newly created accounts. However, certain actors, such as national or international governments, can demand user information, and law enforcement has increasingly used third party subpoenas and warrants to seek personal user information.

Id. at 981-82.  “Similar to Google, Microsoft collects an abundance of personal information through Bing search, and because Windows owns most of the worldwide computer market share, most computer-users have access to Bing.”  Id. at 983.

Ms. Edano’s law review article compares keyword warrants to geofence warrants and tower dump warrants.  A petition to the U.S. Supreme Court is pending on geofence warrants. Certiorari Petition to U.S. Supreme Court in 4th Circuit Geofence Decision – E-Discovery LLC (Aug. 4, 2025).

For a discussion of the Stored Communication Act (“SCA”) and keyword warrants, see Google LLC v. United States, 2025 WL 778150, at *1 (D.D.C. Feb. 25, 2025).

[T]he SCA permits the Government to require the disclosure of the widest possible range of information if it applies for a “warrant” “using the procedures described in the Federal Rules of Criminal Procedure” or a state counterpart. [18 U.S.C.] § 2703(a), (b)(1)(A), (c)(1)(A). SCA warrants require no prior notice to the subscriber, but, under Federal Rule of Criminal Procedure 41, they — like all warrants — require “a judicial finding of probable cause based on sworn testimony or an affidavit.”

Id. at *1.  The Google court wrote:

Throughout its investigation [Redacted], the United States has repeatedly obtained a series of “reverse search” warrants directed to Google for information pertaining to various of its account holders or users. Reverse-search warrants are a peculiar species of warrant. They “differ from warrants authorizing surveillance of a known suspect, since they are normally employed when the identity of a suspect is unknown.” …The Government typically goes through a two-step process when seeking such warrants…. At step one, it obtains a warrant for anonymized information about various accounts or users that match certain criteria — e.g., that they were in a particular location (a “geofence” warrant) or searched for certain terms (a “keyword” warrant) within a given timeframe…. After reviewing and culling that information, the Government then — at step two — obtains a second warrant or court order to de-anonymize accounts it wishes to investigate further.

Id. at *2 (emphasis added).

In Google, the government served a keyword warrant and the court described the process: “In response, Google turned over thousands of responsive searches along with anonymized user data for an estimated 1,341 users’ who had made those searches….  The Government then began a process of reviewing that information and seeking de-anonymization of the user data in tranches — step two. ”  Id.  It then obtained a supplemental warrant for identifying information on users who made ~293 of the searches.  Id.  “Google subsequently provided de-anonymized information on 251 users.”  Id.  The government than obtained a second warrant “for another batch of users”; however, Google had already purged those files. Id.

A year and a half later, the United States procured the warrant at issue here…. Approved again under § 2703 and Rule 41, it sought identifying information about users who had made approximately 537 of the searches previously identified (“querying users”), as well as anyone who had shared devices or identifiers with those users (“technically connected users”)…. For the querying users — thought to encompass between 283–311 accounts — the warrant demanded the same information as did prior supplemental warrants…. Slightly more circumscribed information was sought for the technically connected users.

Id. at *2.

Google then moved to quash, unsuccessfully.  The court held that Google did not have “any statutory or constitutional right to bring a pre-execution motion to quash on the ground that the warrant violates the Fourth Amendment.”  Id. at *3.  Next, it clarified “that, in any event, no changed circumstances call into question the original probable-cause determination, and Google’s invocations of the First Amendment and the Due Process Clause do not entitle it to any further protections.”  Id. Finally, the court concluded “by rejecting Petitioner’s claim that the records requested are too voluminous in nature to produce and that compliance would thus pose an undue burden.”  Id. After explaining its analysis in detail, the Google court wrote: “For the foregoing reasons, the Court will affirm the Order below and deny Google’s Motion to Quash. A separate Order so stating will issue this day.”

“The full extent of law enforcement’s use of keyword search warrants remains largely unknown.”  Brian L. Owsley, “Searching A Person’s Thoughts: Keyword Search Warrants and Fourth Amendment Concerns,” 28 Stan. Tech. L. Rev. 66, 69 (2024).

One example is a 2017 fraud case.  The fraudster obtained a funds transfer using the victim’s identifying information.  The police “sought to compel Google to produce information on all persons or devices who searched for the victim’s name during the 38-day span of December 1, 2016, to January 7, 2017.”  Id. at 73. The warrant was issued and served; however, the results are not known. Id.

In 2018, a bomb in Austin, Texas, killed a man.  It exploded on his front porch.  Additional bombs exploded. “After the first three explosions, the FBI sought and received three keyword search warrants. These warrants sought information from the search engines operated by Google, Yahoo, and Microsoft for various iterations of the street addresses where the explosions occurred.”  Id. at 74.  After a fourth bomb, the FBI applied for information of searches around the explosion site based on street names. Id.  Later, :”the FBI filed a fifth search warrant application directed at Google related to the Austin explosions. Specifically, this warrant requested data from Google for various search terms like ‘pipe bomb,’ ‘PVC bomb,’ ‘explosive,’ ‘ied,’ ‘trigger,’ ‘package,’ and ‘fragmentation.’”  A suspect was identified, fled, and detonated a bomb, killing himself. Id.

Prof. Owsley describes a number of other uses of keyword searches, id., passim; see also Nicole Chan, “Wrong Search at the Wrong Time: Keyword Search Warrants and the Fourth Amendment,” 45 Cardozo L. Rev. 271, 272, 284 (2023), and the technical process in detail:

When Google is served with a keyword search warrant compelling the company to identify users who searched for a particular term during a particular time frame, it implements a staged process to produce a response for law enforcement.

First, an employee in Google’s Legal Investigation Support team formulates a text-based query matching the warrant’s specifications. The query is then run against billions of daily search records from Google Search and Google Maps. (Users can delete their search records, potentially excluding them from these searches.)

Once matches are identified from this search, the Legal Investigation Support team creates a CSV file containing records of searches matching the specified terms. These may be limited to exact matches, or, more commonly, “the results may extend to queries that include the specified search terms as part of a query that contains other words.”

Next, Google de-identifies the CSV results to remove the account-identifying information from the results. Google produces this de-identified version to law enforcement as “potentially responsive data” that includes a number of different categories of information. These various categories include the search date and time; the coarse location data of where the user conducted the search based on the internet protocol (“IP”) address; the user’s actual search; and the user’s actual results from the Google search. Moreover, this initial production would also contain the host, which is the Google domain name that the user contacted. Typically, it would include the request, “which is the latter part of the URL, following the host, that is associated with the user’s search.” When the search is performed by an authenticated user’s account, the results would also contain an abbreviated Google identifier known as a GAIA ID. If the user does not have an authenticated account, then an abbreviated version of that person’s Browser Cookie ID would be provided. Finally, the categories include “the associated user agent string.”

By abbreviating either the GAIA ID or the Browser Cookie ID, the Litigation Investigation Support team de-identifies the data produced in response to the keyword warrant. After receiving this initial production, a law enforcement officer reviews it to determine whether the de-identified results might be relevant to the criminal investigation. For example, Google’s initial de-identified production might reveal a number of users searched for terms requested by the warrant during a specific timeframe. After a review of the de-identified data related to those users, law enforcement officers may exclude searches for which the complete query contained additional search terms outside the parameters of their search. Additionally, officers may exclude results related to addresses in cities or states outside the targeted criminal investigation area.

When the original search warrant does not authorize the law enforcement agency to obtain identifying data regarding the targeted suspects, the officers must apply for another search warrant ordering the disclosure of this data and information. Here, when Google provides information based on abbreviated GAIA IDs, the data includes the IP address related to the search (when available), the full GAIA ID, and the basic subscriber information associated with that GAIA ID, as defined in the Stored Communications Act. For further queries that include abbreviated Browser Cookie IDs instead, Google provides the IP address related to the search as well as the full Browser Cookie ID.

Google maintains that this process is designed to safeguard its users’ privacy, with narrowly-tailored search parameters and a two-stage deanonymization process helping to ensure that law enforcement only gets the  information that would prove useful in investigating a particular crime.

Id. at 69-71; see also Chan, “Wrong Search,” 45 Cardozo L.Rev. at 287.[1]

____

[1] A forthcoming blog, “Geofencing, Geotargeting, and Geo-blocking, in Civil Litigation, Information Governance, and Regulation,” will address how those concepts are beginning to appear in civil contexts.