Deepfake Evidence Rules Explained: Draft 901(c) and Rule 707

The federal judiciary's Advisory Committee on Evidence Rules met in Boston on October 15, 2026 to take expert testimony on deepfake and AI-generated evidence. No Federal Rule of Evidence governs either subject. The rules that decide whether your recording gets admitted are the ordinary authentication rules, applied case by case.

Information last verified on October 18, 2026. This is a developing rulemaking matter and the posture below is stated as of that date.

Status: No federal evidence rule addresses deepfakes or AI-generated evidence. Draft Rule 901(c) is labeled a "Working Draft" that the Advisory Committee has prepared but not finally approved and has never published for public comment. Proposed Rule 707 was published for comment in 2025; the October 2026 agenda says the Committee voted to withdraw the published version and submit a modified proposal for further study, and the Committee's own May 2026 report says it decided not to seek re-publication for now. The Committee met on October 15, 2026, and as of October 18, 2026 no minutes, report or supplement from that meeting has been published.

Jurisdiction scope: This article addresses the Federal Rules of Evidence and the federal rulemaking process for authenticating recordings and other digital items challenged as AI fabrications. It does not address state evidence codes, which vary, or the separate criminal and civil statutes that penalize creating or distributing deepfakes. For those, see our state-by-state survey of deepfake and AI voice cloning laws.

What Happened

The Advisory Committee on Evidence Rules held its fall meeting in Boston on October 15, 2026. Its published agenda set the morning aside for two panels of outside experts, one on deepfakes and one on AI offered as evidence without an accompanying expert witness. The agenda book for that meeting, which the Administrative Office of the U.S. Courts published in advance, runs to 221 numbered pages plus appended rules suggestions.

The agenda itself states the posture of the deepfake proposal directly. "At its last seven meetings, the Committee has been working on a possible amendment to address the evidentiary challenges raised by 'deepfakes,'" it reads, and "The Committee has prepared, but not finally approved, a new Rule 901(c) that would require the opponent to provide evidence sufficient to support a finding that a challenged item was a deepfake; if that happens, the proponent would then have to show that the item is more likely than not authentic."

The Committee was more explicit in its May 17, 2026 report to the Standing Committee. That report states that "at least for now, an amendment to Rule 901 to address deepfakes is not warranted," because "few examples exist of courts having to address the possibility of deepfakes," and describes the objective as fine tuning a possible amendment "to hold in abeyance" until the Committee concludes one is warranted.

The Reporter to the Committee, Professor Daniel J. Capra of Fordham University School of Law, is blunter in his September 15, 2026 memorandum behind Tab II. The Committee has been working on a new Rule 901(c), he writes, "but has, to date, been unsure about the need for such an amendment."

The Rule 707 proposal has traveled further and come back. The Committee approved it for public comment at its spring 2025 meeting by a vote of 8 to 1, with the Department of Justice voting against, according to the Committee's own report in the August 2025 preliminary draft. As published for comment, Rule 707 read: "When machine-generated evidence is offered without an expert witness and would be subject to Rule 702 if testified to by a witness, the court may admit the evidence only if it satisfies the requirements of Rule 702(a)-(d). This rule does not apply to the output of simple scientific instruments."

The comment period closed on February 16, 2026. The Committee held two public hearings and received 59 written comments by the deadline, according to the draft minutes of the May 7, 2026 meeting, though the October 2026 agenda refers to more than 70. The minutes classify 3 of those comments as expressing unqualified support, 27 as supporting the rule subject to revisions and 27 as opposed, a breakdown that accounts for 57 of the 59. The October 2026 agenda records what happened next: "the Committee at the last meeting voted to withdraw the proposed rule as issued for public comment, and to submit a modified rule proposal for further study."

The Committee's own report to the Standing Committee, dated May 17, 2026, describes the same decision differently. It states that the Committee "does not recommend action on the proposed Rule 707 at this time" and that it "decided not to propose that the revised Rule 707 be released for a new period of public comment at this time." The draft minutes of the May 7 meeting record no vote to withdraw, only unanimous agreement that discussion continue at the fall mini-conference.

That was an unusual step, and the Committee's Reporter said so. At the Standing Committee's June 3 and 4, 2026 meeting, according to its minutes, "Professor Capra added that this was the first time in his experience that a rule published for public comment did not go forward."

The official pipeline confirms where both proposals now sit, which is nowhere. The only Evidence Rules currently published for public comment are Rules 104 and 902, open from August 14, 2026 to February 15, 2027. The only Evidence Rule awaiting a December 1, 2026 effective date is Rule 801. Neither Rule 707 nor Rule 901(c) appears at any step of the Rules Enabling Act process.

What the October 15 meeting did is not public

We looked for a record of the outcome and found none. The Committee's October 2026 agenda-book page on uscourts.gov links only the agenda book itself. The agenda-books index lists no Evidence Rules materials later than October 2026 and no January 2027 Standing Committee book. The published-for-comment and pending-amendments pages were both unchanged as to Rules 707 and 901(c).

As of October 18, 2026, no minutes, report, supplement or other official record of what the Advisory Committee on Evidence Rules did on October 15, 2026 has been published. That is normal: draft minutes of a fall meeting ordinarily appear in the following spring's agenda book, and any recommendation would surface in the Standing Committee's January materials. It also means nothing can be said about the outcome. The agenda describes what the Committee planned to discuss, not what it resolved.

What the Law Actually Says

Strip away the rulemaking and a reader's question is concrete. The other side says your recording is AI. What do you have to prove?

Rule 901(a) sets a low bar, and that is the whole point of the dispute. It provides that "[t]o satisfy the requirement of authenticating or identifying an item of evidence, the proponent must produce evidence sufficient to support a finding that the item is what the proponent claims it is." That is a sufficiency standard, not a persuasion standard. The judge asks whether a reasonable juror could find the item genuine, not whether the judge believes it is.

Rule 901(b) lists illustrations, and two of them carry most audio and video. Rule 901(b)(1) accepts "[t]estimony that an item is what it is claimed to be" from a witness with knowledge. If you recorded the call, you can say so. Rule 901(b)(9) accepts "[e]vidence describing a process or system and showing that it produces an accurate result," which is the usual route for surveillance footage, dashcam files and automated logs. Rule 901(b)(5) allows voice identification "based on hearing the voice at any time under circumstances that connect it with the alleged speaker." The list is expressly "examples only" and "not a complete list."

Rule 902 lets some items authenticate themselves. Rule 902(13) covers "[a] record generated by an electronic process or system that produces an accurate result, as shown by a certification of a qualified person," and Rule 902(14) covers data copied from a device and authenticated by digital identification. Both require the notice procedure of Rule 902(11). Neither asks anyone to rule out fabrication.

Rule 104 allocates the decision. Under Rule 104(a), "[t]he court must decide any preliminary question about whether a witness is qualified, a privilege exists, or evidence is admissible," and in doing so is not bound by the evidence rules except those on privilege. Under Rule 104(b), when relevance depends on a fact, "proof must be introduced sufficient to support a finding that the fact does exist." Authenticity has historically been handled on that lower, sufficient-to-support-a-finding footing.

Rule 702 is the reliability gate, and it is the hinge of Rule 707. Rule 702 allows expert opinion only "if the proponent demonstrates to the court that it is more likely than not" that the four conditions in subdivisions (a) through (d) are met. The gap Rule 707 was drafted to close is that Rule 702 governs humans. When an AI system produces the inference and no expert testifies to it, there is no obvious rule demanding that the system be shown reliable.

So, practically: a party objecting that a recording is a deepfake is making an authenticity objection under a rule that was not written with synthetic media in mind. If the proponent produces a witness who says "this is my recording," Rule 901(a) is ordinarily satisfied, and the fabrication argument becomes an argument about weight for the jury rather than admissibility for the judge. That allocation is what the draft rule would change. Our page on how US courts treat recordings offered as evidence walks through how the analogous state-court rules play out.

Courts are not powerless in the meantime, and the draft Committee Note says so. It points to Rule 403 as a tool against a generalized "you cannot believe anything you see" presentation, citing United States v. Peterson, 945 F.3d 144, 157 (4th Cir. 2019), which the Note describes as upholding exclusion of a demonstration of how easy it is to fake a text message. It also points to a court's authority over lawyer argument, citing Lee v. City of Troy, 339 F.R.D. 346, 367 to 368 (N.D.N.Y. 2021), which the Note describes as a case where defense counsel argued without any basis that the plaintiff's videos were "manufactured." And the Reporter's memorandum collects examples of courts punishing fabrication directly, including a Kentucky appellate decision the memo describes as upholding a 45-day contempt sentence against a party who, in a family law timesharing dispute, ultimately conceded that her testimony and the documents she had submitted to the court were false or altered.

What Draft Rule 901(c) Would Change

The working draft printed at page 150 of the October 2026 agenda book, with its draft Committee Note running through page 152, is titled "Potentially Fabricated Evidence Created By Generative Artificial Intelligence." It has five subdivisions.

Subdivision (1) sets the trigger. If a party challenges authenticity "on the ground that it has been fabricated, in whole or in part, by generative artificial intelligence, the party must present evidence sufficient to support a finding of fabrication to warrant an inquiry by the court." Subdivision (2) then shifts the burden: "If the opponent makes that showing, the item of evidence is admissible only if the proponent demonstrates to the court that it is more likely than not authentic."

Subdivision (3) applies the rule to items offered under either Rule 901 or Rule 902. Subdivision (4) requires that "[u]nless the court orders otherwise, a party challenging an item of evidence under this subdivision (c) must provide reasonable pretrial notice of the intent to present evidence of fabrication, so that all other parties have a reasonable opportunity to respond to that evidence." Subdivision (5) defines "generative artificial intelligence" as "a computer system that emulates the structure and characteristics of input data in order to create derived synthetic content, including images, video, audio, text, and other digital content," language the Note says is derived from Executive Order 14110 of October 30, 2023.

The draft Committee Note is explicit that the change would be real rather than cosmetic. It states that once the opponent meets the threshold, "the proponent will not establish authenticity simply by satisfying one of the illustrations of authenticity set forth in Rule 901(b)." Testimony of a witness with knowledge under Rule 901(b)(1) "will provide evidence sufficient to support a finding that the item is authentic under Rule 901(a)," the Note says, "[b]ut that knowledge will not ordinarily prove more likely than not that the item is authentic, especially given the opponent’s submission showing some likelihood of a deepfake." The justification the Note gives is that "a person with 'knowledge' may well think that the item is genuine when in fact it is a deepfake."

The Note is equally explicit that a bare accusation would not be enough. "[A] broad claim of 'deepfake' is not enough to put the court and the proponent to the time and expense of showing that the item has not been manipulated by generative artificial intelligence," it reads, and "[a] proponent should not in the first instance be required to show that every video or audio it proffers is not a deepfake."

The Reporter's memorandum proposes adding a non-exhaustive list of facts pertinent to the challenger's threshold showing: missing or anomalous metadata; inconsistent timestamps or creation dates; unexplained edits or compression artifacts; absence of the native file when a derivative version is offered; gaps in chain of custody that would raise an opportunity for tampering; glitches such as unusual audio patterns and problems of synchronization; known use of generative AI tools by the proponent; and independent evidence inconsistent with the content of the item. The memorandum also flags the risk in publishing any such list, noting that generative AI "is moving at lightspeed, and so factors that raise suspicion right now might never even occur in three years."

Two carve-outs matter. The Note states the rule "does not apply if the preparation and issuance of a deepfakes is the gravamen of the claim or charge," and separately that it does not apply "when a proponent offers a deepfaked item for the purpose of proving that it is fabricated evidence."

Where Rule 707 Now Stands

The version the Committee tentatively approved in May 2026, printed at pages 173 and 174 of the agenda book, is retitled "Evidence Produced By Artificial Intelligence and Presented at Trial Without an Expert." Subdivision (a) carries the four Rule 702 reliability conditions directly into the rule text rather than cross-referencing them. Subdivision (b) is the part the Committee fought over: "Ordinarily, the proponent must provide an expert to explain how the system of artificial intelligence reliably produced the evidence. But in exceptional circumstances, the court may consider other proof to find that the output satisfies the requirements of (a)(2)-(4)." Subdivision (c) adds notice, subdivision (d) carves out facts subject to judicial notice under Rule 201, and subdivision (e) defines "artificial intelligence."

The Reporter's memorandum is careful about what that approval means. "This is not to say that the Committee has approved all of the language in the working draft or the Note," it states. "Rather, the draft rule and note take an approach that was approved by a majority of the Committee for situations in which AI is admitted without an expert to prove a fact."

The Department of Justice opposed both drafts. Elizabeth J. Shapiro, Deputy Director of the Federal Programs Branch of the DOJ Civil Division, told the Committee in May 2026 that "the DOJ does not believe Rule 901(c) is ready for publication," according to the draft minutes. She said the notice provision's "unless the court orders otherwise" clause would be routinely excused for criminal defendants unwilling to share trial strategy; that the generative-AI definition would be hard to apply where evidence is created through a combination of generative AI and other AI functions; and that a defendant might try to use Rule 901(c) against government-offered AI-generated images in child exploitation cases. The Reporter responded that the rule would not reach images the government concedes are fabricated, because that is not a deepfake offered as something it is not. The minutes record Shapiro's position that the existing Federal Rules of Evidence "are adequate to address the issue of potential deepfake evidence," and that "the Committee has much more work to do before proposing a deepfake rule." On Rule 707, the minutes record that she "stated that the DOJ would rather have no Rule 707 at all."

Other participants pushed the other way. The minutes record the Federal Defenders' representative stating that a rule covering AI-generated evidence offered without an expert "is not premature" and that federal defenders are already seeing such evidence in cases.

The minutes also record why no vote was taken on publishing Rule 901(c). The chair, Judge Jesse M. Furman of the Southern District of New York, concluded that in light of the disagreement on the Committee, and notwithstanding a delay of another year, the better course was a mini-conference at the fall 2026 meeting, and "stated that there would be no need for a Committee vote regarding public comment now given that plan." On Rule 707, the minutes record that "Committee members then unanimously agreed that discussions regarding Rule 707 should continue at the Fall 2026 mini-conference with a focus on Version 2," and that Judge Furman "noted that combining the consideration of Rules 707 and 901(c) would create helpful synergy and would allow for potential publication of both proposed AI rules together."

Analysis: Why This Matters

The following is analysis from the Recording Law Editorial Team.

The evidence here is mostly about scarcity. The Federal Judicial Center surveyed all federal trial judges at the Committee's request. Per the draft May 2026 minutes, the response rate was 45 percent, 931 judges responded, and 15 reported having encountered a deepfake challenge. Among those who had not, 82 percent said they would require some initial showing before entertaining a fabrication claim, which is substantially what the draft rule would codify, and a narrow majority said the rules should be amended.

Those two findings pull against each other, and that tension explains the deadlock on the Committee better than any single position does. One reading is that the problem has not arrived in federal court and a rule would regulate a hypothetical. Another is that judges are already applying the draft rule's logic without a rule, which argues the draft describes sound practice rather than inventing it. Both readings appear in the minutes, and the Committee has not resolved them.

What is worth noticing is that the burden-shift, not the deepfake vocabulary, is the operative change. Rule 901(a)'s sufficiency standard is deliberately permissive because authenticity disputes were historically thought to be jury questions. Draft Rule 901(c) would move a subset of those disputes to the judge under a preponderance standard. That is a structural change to who decides, and it would apply to a category defined by how the challenger frames the objection. Both the DOJ and at least one Committee member raised the asymmetry that follows, and the Committee did not resolve it.

The second thing worth noticing is the timeline, because it sets a floor on how long the current law stays current. Judge Furman observed in May 2026 that Rule 901(c) could not take effect before December 1, 2028 even on a summer 2026 publication. No publication happened. Whatever the Committee did on October 15, 2026, the ordinary authentication rules govern deepfake challenges in federal court for years, not months.

For readers who make and rely on recordings, the practical consequence is narrow but real. "They will say it is AI" is now a live objection, and under current law it is an argument about weight that the other side gets to make to the jury after the recording is admitted. Our coverage of the ELVIS Act and the states that followed it tracks the separate question of when synthesizing someone's voice is itself unlawful, which is a different problem from authenticating a real recording.

How This Affects You

If you are offering a recording in federal court, nothing has changed. The proponent's burden remains the Rule 901(a) sufficiency showing, and the routine routes remain Rule 901(b)(1) testimony from someone with knowledge and Rule 901(b)(9) evidence about the process or system that produced the file. The Reporter's memorandum describes an Eighth Circuit decision in which the court of appeals found a sufficient authentication showing under Rule 901(a) for surveillance video with unexplained gaps, and concluded that the gaps went to weight and not admissibility.

If someone challenges your recording as AI-generated, the current rules do not require you to prove a negative, and no federal rule requires the challenger to make a threshold showing either. How a particular judge handles the dispute is a Rule 104(a) question committed to that judge's discretion, and outcomes vary.

The general practices that help under current law are the same ones that would help if draft Rule 901(c) were ever adopted, because the draft's own factor list is a list of record-keeping failures. Preserving the original native file rather than a re-encoded copy, keeping metadata intact, and being able to describe how the recording was made and stored all speak to authenticity under Rule 901(b)(9) today. None of that is legal advice about your situation, and none of it guarantees admission.

If your matter is in state court, these rules do not apply to it directly. State evidence codes govern, and they differ. The Committee's own minutes note that the Federal Rules of Evidence serve as a model for state courts, and that deepfake issues may start to appear more in state courts, in family court cases for example, than in federal litigation, but a state court is not bound by a federal draft that has not been adopted.

What Happens Next

Draft Rule 901(c) and the revised Rule 707 would each have to clear every step of the Rules Enabling Act process. According to the Administrative Office's description of how the rulemaking process works, an advisory committee evaluates a proposal first; if it pursues the proposal, it asks the Committee on Rules of Practice and Procedure, known as the Standing Committee, for permission to publish a draft for public comment. After the comment period, the advisory committee may discard, revise, or transmit the amendment. The Standing Committee reviews the advisory committee's findings and, if satisfied, recommends the change to the Judicial Conference of the United States, which in turn recommends it to the Supreme Court. If the Court concurs, it promulgates the rule by order "before May 1, to take effect no earlier than December 1 of the same year," and the rule takes effect unless Congress enacts legislation to reject, modify, or defer it.

Neither proposal has completed step one. Rule 901(c) has never been published for comment. Rule 707 was published, and the Committee did not carry it forward after the comment period.

That process is also reversible late in its course, which is not hypothetical here. The Committee's proposed one-word amendment to Rule 609(a)(1)(B) was approved by the Standing Committee in June 2026, and then, as the September 2026 report to the Judicial Conference records, that approval was withdrawn in August 2026 after the Judicial Conference's Executive Committee raised concerns about the committee note, sending the proposal back to the Advisory Committee.

The specific events that would turn this explainer into hard news are identifiable:

• Publication of the October 15, 2026 minutes, ordinarily in the spring 2027 Evidence Rules agenda book, which would show what the Committee did with the two drafts after hearing the panels.
• A Standing Committee agenda book or report, ordinarily published ahead of its January meeting, carrying an Evidence Rules request to publish either draft for public comment.
• An entry for Rule 901(c) or Rule 707 on the Administrative Office's list of proposed amendments published for public comment. That is the first public, verifiable step, and it would open a comment period of roughly six months.
• A Judicial Conference recommendation and a Supreme Court order transmitting a rule to Congress before May 1 of some future year, for a December 1 effective date.

We are not predicting which, if any, of those will happen, in what form, or when. The Committee has had a deepfake amendment on its agenda since 2023 and has not voted to publish one.

This is general legal information, not legal advice. This article addresses the Federal Rules of Evidence and federal rulemaking on deepfake and AI-generated evidence, not any state's evidence code and not the separate statutes that criminalize making or distributing deepfakes. Sources were verified on October 18, 2026. Laws change; consult a lawyer licensed in your jurisdiction about your specific situation.

Related articles

• Deepfake and AI voice cloning laws, state by state
• Deepfake fraud and impersonation statutes, federal and state
• What California requires of AI images and cloned voices

---

Last updated: 2026-10-18. Developing rulemaking matter: no record of the Advisory Committee's October 15, 2026 action had been published as of this date.

Frequently Asked Questions

Is there a federal rule of evidence for deepfakes or AI-generated evidence?

No. As of October 18, 2026 there is no Federal Rule of Evidence addressing deepfakes or AI-generated evidence. Draft Rule 901(c) is a working draft that the Advisory Committee on Evidence Rules has prepared but not finally approved, and it has never been published for public comment. Proposed Rule 707 was published for comment in August 2025, and after the comment period the Committee did not carry the published version forward. The October 2026 agenda describes that as a vote to withdraw the published rule and submit a modified proposal for further study. The only Evidence Rules currently out for public comment are Rules 104 and 902.

What happens today if the other side claims my recording is a deepfake?

The dispute is handled under the existing authentication rules. Under Rule 901(a) the party offering the recording must produce evidence sufficient to support a finding that the item is what that party claims it is, which is a low threshold. Testimony from a witness with knowledge under Rule 901(b)(1), or evidence about the process or system that produced the file under Rule 901(b)(9), ordinarily satisfies it. If that showing is made, a fabrication argument usually goes to the weight the jury gives the recording rather than to whether the judge admits it. The judge decides admissibility under Rule 104(a).

What would draft Rule 901(c) actually do?

It would create a two-step burden shift for items challenged as fabricated by generative artificial intelligence. Under the working draft, the challenger must first present evidence sufficient to support a finding of fabrication to warrant an inquiry by the court. If the challenger makes that showing, the item is admissible only if the proponent demonstrates to the court that it is more likely than not authentic. The draft also applies to items offered under Rule 902, requires reasonable pretrial notice unless the court orders otherwise, and defines generative artificial intelligence. The draft Committee Note states that once the threshold is met, satisfying a Rule 901(b) illustration alone would no longer establish authenticity.

Was proposed Rule 707 adopted?

No. Rule 707 was approved for publication by the Advisory Committee in spring 2025 by a vote of 8 to 1, with the Department of Justice voting against, and the comment period ran until February 16, 2026. The draft minutes of the May 7, 2026 meeting record 59 written comments received by the deadline and two public hearings; the October 2026 agenda refers to more than 70 written comments. The October 2026 agenda states that the Committee then voted to withdraw the proposed rule as issued for public comment and to submit a modified rule proposal for further study. Rule 707 does not appear at any step of the current rulemaking pipeline.

What did the Advisory Committee decide at its October 15, 2026 meeting?

That is not public. The Committee met in Boston on October 15, 2026, and its agenda scheduled two panels of outside experts for that morning. As of October 18, 2026 no minutes, report, supplement or other official record of the meeting had been published. The agenda book describes what the Committee planned to discuss, not what it resolved. Draft minutes of a fall meeting ordinarily appear in the following spring's agenda book.

Does Rule 902 self-authentication protect a recording against a deepfake claim?

Not in the way people sometimes assume. Rule 902(13) makes a record generated by an electronic process or system self-authenticating when a qualified person certifies that the process produces an accurate result, and Rule 902(14) covers data copied from a device and authenticated by digital identification. Both address authenticity, not reliability, and both require the notice procedure of Rule 902(11). Draft Rule 901(c) would expressly apply to items offered under Rule 902 as well as Rule 901.

How soon could a federal deepfake evidence rule take effect?

Not quickly. A proposal must be published for public comment with Standing Committee permission, reviewed again by the advisory committee, approved by the Standing Committee, recommended by the Judicial Conference, and promulgated by the Supreme Court by order before May 1 to take effect no earlier than December 1 of the same year, subject to congressional action. Neither draft has been published for comment. The Committee chair noted in May 2026 that Rule 901(c) could not have taken effect before December 1, 2028 even if it had been published that summer.

How common are deepfake challenges in federal court?

Rare so far, according to the Committee's own data. The Federal Judicial Center surveyed all federal district, magistrate and bankruptcy judges at the Committee's request. The draft minutes of the May 7, 2026 meeting record a 45 percent response rate, 931 judges responding, and 15 reporting that they had encountered a deepfake challenge. Of the respondents who had not encountered one, 82 percent said they would require some initial showing before entertaining a fabrication claim, and a narrow majority said the rules should be amended.

Do state courts have to follow the Federal Rules of Evidence on this?

No. The Federal Rules of Evidence govern proceedings in federal court. State courts apply their own evidence codes, which vary, although many are modeled on the federal rules. The Committee's May 2026 draft minutes record the observation that the Federal Rules serve as a model for state courts and that deepfake issues may start to appear more in state courts, in family court cases for example, than in federal litigation. A federal draft that has not been adopted binds no court.