By Bill Hartzer · Published · Updated · How this site is researched
What happened
Proposed Federal Rule of Evidence 707 addresses machine-generated evidence offered without a sponsoring expert. The version published for public comment 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." Rule 702 is the federal rule that sets the reliability requirements for expert testimony.
According to reporting by Proskauer in the National Law Review, the Advisory Committee on Evidence Rules voted 8-1 on May 2, 2025 to publish the rule, with the Department of Justice dissenting. The Standing Committee (the Committee on Rules of Practice and Procedure) approved publication on June 10, 2025. The comment period ran from August 15, 2025 to February 16, 2026.
Separately, the Advisory Committee had spent several meetings on a possible Rule 901(c), a draft authentication provision for alleged deepfakes (fabricated audio, video or images). As summarized in the committee's report, the draft used a 2-step process. The opponent would first have to show that fabrication is likely, and the proponent would then have to establish authenticity by a preponderance of the evidence.
What was issued or decided
The Advisory Committee met on May 7, 2026. Its report to the Standing Committee, dated May 17, 2026 and signed by Chair Jesse M. Furman, states that the committee "does not recommend action on the proposed Rule 707 at this time. Instead, it has revised the proposed Rule and plans to conduct further study." The report tallies the comments: 3 in unqualified support, 27 supporting the rule subject to revisions and 27 opposed. It lists as criticisms the breadth of the term "machine-generated evidence" and the vagueness of the simple-scientific-instruments exception.
The revised draft, retitled "Evidence Produced by Artificial Intelligence and Presented at Trial Without an Expert," replaces "machine-generated" with "artificial intelligence." It lists the four Rule 702 requirements the proponent must establish, says admissibility "ordinarily requires the proponent to provide an expert," adds a notice provision, replaces the simple-scientific-instruments exception with an exception for facts that may be judicially noticed under Rule 201, and defines artificial intelligence using the National Artificial Intelligence Initiative Act of 2020. The committee agreed that the revised rule would require re-publication. It decided instead to have the draft vetted at its fall 2026 meeting by technology experts and others in the field.
On deepfakes, the report states: "The Committee is of the view that, at least for now, an amendment to Rule 901 to address deepfakes is not warranted." It cites a Federal Judicial Center survey in which 15 judges reported having dealt with deepfake issues, and it plans a mini-conference on the subject at the fall 2026 meeting.
Where it stands
Proskauer reports that the Standing Committee, at its June 2026 meeting, did not recommend action on Rule 707 and returned it for revision and further study. The package of evidence amendments that the federal judiciary published for comment on August 14, 2026 covers Rules 104 and 902 only. Neither Rule 707 nor a Rule 901(c) is in it. As of September 11, 2026, no federal rule of evidence specific to AI output or deepfakes is in effect or pending final approval.
Update, October 2026. The agenda book for the Advisory Committee's October 15, 2026 meeting in Boston confirms that posture in the Committee's own words. After two hearings and "more than 70 written comments," the Committee "voted to withdraw the proposed rule as issued for public comment, and to submit a modified rule proposal for further study." On deepfakes, it "has prepared, but not finally approved, a new Rule 901(c)" that would require the opponent to offer evidence sufficient to support a finding that an item is a deepfake, after which the proponent would have to show the item is more likely than not authentic. The meeting opens with expert panels on both subjects, introduced by the Reporter's memoranda of September 15, 2026. The deepfake memorandum's survey of new cases includes Adams v. Anderson, a Kentucky contempt case over evidence alleged to be AI-fabricated. As of October 8, 2026, neither rule has been approved for adoption.
Update, October 21. The Advisory Committee met in Boston on October 15 and heard expert panels on deepfakes and on AI-generated evidence. The speakers from OpenAI and Meta sat on the deepfake panel; the Rule 707 panel included speakers from Princeton, NIST, the Justice Department and the federal defenders. According to press reports, Judge Jesse M. Furman, the committee's chair, appointed a subcommittee to examine the deepfake question further. No vote to publish a revised Rule 707 or a Rule 901(c) was reported, and the draft minutes of the May meeting, printed in the agenda book, record the Justice Department's view that Rule 901(c) was not ready for publication. The October minutes were not yet posted when this entry was checked.
Update, October 30. The October minutes were still not posted on uscourts.gov, and no vote on either rule has been reported. The published 2025 text and its Committee Note were checked against the judiciary's own preliminary draft. On the committee calendar, the next Advisory Committee meeting is April 14, 2027. If a revised Rule 707 were approved for republication then, the ordinary sequence of comment, approval and Supreme Court adoption would make December 1, 2029 the earliest effective date; that is this record's calculation, not a date the Committee has announced. The full text of both drafts, the public comment and the state counterparts are set out in the guide Proposed Rule 707: AI Evidence and the Daubert Standard.
Update, November 2. The October minutes were still not located on uscourts.gov, and no vote on either rule has been reported. Press accounts of the deepfake panel add detail the agenda book does not carry. According to IBTimes UK, Shira Anderson, policy counsel at OpenAI, told the committee she did not expect "a scientific way going forward to detect" AI-generated images, and Devin Norder of Meta said media from recent models can be indistinguishable from authentic material to a human viewer. The same report says Judge Valerie Caproni, a committee member, cautioned against an AI-specific rule without stronger evidence that deepfakes are a widespread courtroom problem, and that the Federal Judicial Center survey the committee relies on found 15 of 931 responding judges had faced a deepfake challenge. Reuters also covered the session on October 16. None of this changes the posture: Rule 707 remains a revised draft under study, and Rule 901(c) remains prepared but not approved.
How it maps to the Standard
The proposal engages Clause 4 of the Standard, checkability before adoption. Its premise is that output from a system, offered without a person to vouch for it, should meet the same reliability showing as an expert's opinion. An expert who relies on such output in a report is already exposed to Rule 702 scrutiny. An expert should therefore be able to show how the output was checked and on what data.
It also engages Clause 5, reproducible analysis. A method that can be re-run and explained is the kind of evidence a reliability inquiry can test. The revised draft's notice provision points in the same direction as Clause 8: the opposing party learns that AI was used before trial.
Sources
- Advisory Committee on Evidence Rules (uscourts.gov) · May 2026
Report of the Advisory Committee on Evidence Rules - United States Courts · August 15, 2025
Preliminary Draft of Proposed Amendments (August 2025), including proposed Rule 707 and Committee Note - Committee on Rules of Practice and Procedure · June 3, 2026
Agenda book, June 3-4, 2026 meeting (Advisory Committee report dated May 17, 2026) - United States Courts · August 14, 2026
Proposed Amendments Published for Public Comment - National Law Review (Proskauer) · July 29, 2026
Machine-Generated Evidence Challenges the Federal Rules: Inside the Contested Proposed Federal Rule of Evidence 707 - Nelson Mullins · June 11, 2025
Safeguarding the Courtroom from AI-Generated Evidence: Federal Rule of Evidence 707 - Advisory Committee on Evidence Rules · September 23, 2026
Agenda book, October 15, 2026 meeting (Boston) - Court Cast · October 16, 2026
AI deepfakes put courtroom evidence under growing cloud of doubt, OpenAI and Meta warn - IBTimes UK · October 16, 2026
OpenAI Warns Courts May Have No Reliable Way to Spot AI Deepfake Evidence
What could not be confirmed
The deepfake subcommittee reported after the October 15 meeting appears only in press reports (IBTimes UK and Court Cast); no member is named and no primary record of it was located as of October 30, 2026. Comment counts differ across sources: the reporter's summary counts 59 comments addressing Rule 707 (3, 27 and 27 by position, plus others not classified); the October agenda book and Baker Botts report more than 70 written comments, which likely include comments on other rules. The summary says 13 witnesses testified; the two transcripts name more speakers, some of whom addressed other rules. The published 2025 text was confirmed on uscourts.gov on October 30, 2026. As of November 2, 2026: the October minutes were not located; speaker statements and Judge Caproni's remarks rest on IBTimes UK, not a transcript; the Reuters report of October 16, 2026 was identified through the OECD AI Incidents Monitor's source list and was not opened; the 931 denominator for the Federal Judicial Center survey comes from press reports, while the committee's May report gives only the figure 15.
About this entry
This entry was checked against the primary document listed in the sources. The entry for Proposed Federal Rule of Evidence 707 records conduct and what the tribunal did. It does not characterize a holding and it is not legal advice. Last checked November 2, 2026. More matters like this one are collected under the rules and guidance entries. If something here is out of date, write to [email protected].