Machine-Generated Evidence: Rules Advisory Committee Hits Pause to Consider Significant Revisions to Proposed Federal Rule of Evidence 707 Addressing Evidence Generated by Artificial Intelligence

On May 7, 2026, the Advisory Committee on Evidence Rules voted not to advance the original Rule 707 proposal or any of the three redrafts the Reporter had circulated for potential republication.  

Instead, the Committee will convene a Fall 2026 mini-conference of technical experts and practitioners (the first portion of its scheduled October 15, 2026, meeting in Boston, in symposium format) to test the viability of the “leading” redraft that was published alongside the related AI “deepfakes” proposal (Rule 901(c)).  The following summarizes the originally proposed rule, the principal public-comment critiques, the redrafts considered by the Committee, the Committee’s May 7 decision, and the revised path forward.

*****

Proposed Federal Rule of Evidence 707, published for comment in August of 2025, would have applied Rule 702 reliability requirements to machine-generated output.  After publication, the draft rule (and its accompanying proposed Advisory Committee Note) received significant criticism from both sides of the bar, as well as from academics, in two public hearings and in written submissions.  In the lead up to the May Advisory Committee meeting, the Committee Reporter published a memorandum titled “Re: Artificial Intelligence and Proposed Rule 707” (“Reporter’s Memo”) (April 2026) as part of the meeting Agenda Book.  The Reporter’s Memo (1) summarized public comments on the original proposed rule; (2) summarized testimony given at the Committee’s public hearings on January 15 and 29, 2026; (3) discussed an alternative proposed rule that was submitted via public comment; (4) set forth three potential redrafts of the proposed rule; and (5) suggested next steps for the Committee.  At the May 7, 2026, meeting in Washington, D.C., the Committee declined to advance the proposed Rule 707 any further in the process and also declined to substitute any of the proposed redrafts for republication.  Instead, the Committee directed that the issue be revisited at a Fall 2026 mini-conference adjoined to the next regularly scheduled Advisory Committee meeting in Boston, Massachusetts on October 15, 2026.

What Was Pulled from the Table and Why

The original August 2025 proposed text for Rule 707 is short and targeted:

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.

Although this proposed text did not require expert testimony, the corresponding proposed Committee Note acknowledges that meeting the 702(a)-(d) requirements would be challenging and potentially unachievable without expert testimony.

The Reporter’s Memo identified seven primary critiques of the August proposed rule that arose via public comment or at the two hearings:

1. The proposed rule is premature given the limited body of reported AI-evidence case law (i.e., AI technology and case law are evolving rapidly, which risks uninformed rulemaking).
2. Existing rules (e.g., 702, 901, 902, and 403) already regulate AI-generated evidence.
3. “Machine-generated” language is overbroad, sweeping in routine computer outputs long handled as business records.
4. The “simple scientific instruments” carve-out is vague and unworkable.
5. The proposed rule needs a notice provision (i.e., parties and courts need notice of intent to use AI outputs as evidence to proactively address potential issues).
6. The choice to tie AI reliability to Rule 702 rather than set freestanding reliability requirements. 
7. The absence of a mandatory expert-foundation requirement in the proposed rule.

New Alternative Draft Rules Surface

In response to the critiques, the Reporter’s Memo offered three potential redrafts of Rule 707.  All three replaced “machine-generated” with “artificial intelligence,” deleted the “simple scientific instruments” exception, added a notice provision, restated the elements of Rule 702(a)–(d), and added a judicial-notice carve-out to Rule 201.  The redrafts differed on expert foundation: Redraft 1 required an expert at the Rule 104(a) foundation hearing only; Redraft 2 required an expert at trial, with an exceptional-circumstances exception; and Redraft 3 required an expert at both foundation and trial.

None of the redrafts resolved the potential overbreadth of any rule as applied for AI tools integrated into everyday business systems or the threshold concern that the AI-as-expert-testimony premise would open a path for parties to offer machine output as a substitute for a human expert.  The Reporter’s Memo also set forth an alternative draft proposed by Jonathan M. Redgrave, Benjamin M. Redgrave, and Timothy P. Redgrave in public comment (USC-RULES-EV-2025-0034-0061 (Feb. 16, 2026)), along with the Reporter’s critique of the alternative formulation.

The May 7, 2026, Committee Decision

At its May 7, 2026, meeting, the Advisory Committee voted not to advance the original Rule 707 proposal or any of the three proffered redrafts for republication.  The companion Rule 901(c) deepfakes proposal was likewise postponed and will not proceed at this time for publication.  The Committee will instead convene a mini-conference of technical experts and practitioners at the first portion of its October 15, 2026, meeting in Boston, in symposium format, to test the viability of the leading Rule 707 redraft alongside the deepfakes proposal.

While the discussion was not extensive, it was clear that the sentiments reflected in the Reporter’s Memo made it infeasible to proceed with the original Rule 707 draft or any of the three potential alternative drafts.  A committee researcher, the Chair, and the Reporter have indicated that “Redraft Two” (which ordinarily requires the proponent to provide an expert witness to explain how the AI system reliably produced the evidence, with an “exceptional circumstances” exception for other proof) is the leading contender.  The Committee has signaled that some respondents to the original public-comment process may be invited to present at the mini-conference.

It also appears that several factors drove the Committee’s caution as it relates to deepfakes.  The Federal Judicial Center surveyed every federal trial judge; more than 930 responded, and only 15 reported encountering a deepfake challenge.  Among those 15, one-third supported amending the authenticity rules.  Among the roughly 900 judges who had not encountered deepfake challenges, 55% supported amendments.  On the record, DOJ representative Betsy Shapiro said the deepfakes rule is “not ready for publication”; Judge Valerie Caproni was “sympathetic,” noting the Committee is “trying to write a rule for a situation that hasn’t come up much”; and Judge Richard Sullivan was “uncomfortable” with floating rules “not really thought that carefully about.”

That said, the Reporter’s Memo reflects the Committee’s tentative approval of a new Rule 901(c) governing “Potentially Fabricated Evidence Created by Generative Artificial Intelligence.”  The proposal uses a two-step framework: a party challenging authenticity must first produce evidence sufficient to support a finding of fabrication, and only then must the proponent demonstrate that the item is “more likely than not” authentic (lifting authentication from Rule 104(b)’s permissive standard to a Rule 104(a) preponderance).  The updated draft also adds subdivisions for pretrial notice and a definition of “generative artificial intelligence.”  Substantively, the proposal appears to remain intact although the Fall mini-conference will provide a chance to revisit draft Rule 901(c) prior to publication.

In the end, the Committee’s posture going forward is deliberative.  Standing Committee Chair Judge James Dever framed deliberation as “a feature, not a bug” of the process.  With republication delayed pending the Fall mini-conference, the earliest publication of any AI evidence amendment for public comment is now August 2027. 

Conclusion

The Committee’s decision to pause both Rule 707 and Rule 901(c) proposals pending a Fall 2026 mini-conference creates an unusual opportunity for additional substantive input.  Corporate counsel and AI-policy stakeholders should track who is invited to the October 15, 2026, Boston symposium and consider whether to participate in or submit material in advance of the meeting.  In the interim, it is clear that courts across the country will continue to apply the Rules of Evidence “as is” to the circumstances presented to make rulings.  In addition to the expected tide of decisional law, local rules or chambers’ guidance may also surface.  Clients and counsel will need to carefully monitor these developments in each jurisdiction.

Redgrave LLP’s Perspective

Redgrave LLP monitors developments at the intersection of evidence law, AI governance, and eDiscovery practice, advising corporate clients and counsel regarding evidentiary issues involving AI-generated data and documents.  We also track and participate in federal, state, and local rulemaking concerning evidentiary and civil procedure issues, including those addressing AI.

For additional information on this topic, please reach out to your Redgrave LLP contact, or you can connect with Robert Keeling, Jonathan Redgrave, Erica Zolner, and Gareth Evans.

Redgrave LLP provides this information for educational purposes only. It should not be construed or relied on as legal advice or to create a lawyer-client relationship.