home  /  insights  /  who-vouches-for-machine-generated-evidence
AI Law & Rules

Does machine-generated evidence need a human expert to sponsor it?

The proposed Federal Rule that would have required one stalled in June 2026 and went back to the Advisory Committee, together with the deepfake-authentication proposal. That leaves the question to Rule 702, Rule 901 and the judge in front of you.

September 4, 2026 · 4 min read

The short answer

As a matter of federal rule text, no — because the rule that would have required it has not been adopted. Proposed Federal Rule of Evidence 707 would have subjected machine-generated evidence offered without a human expert to Rule 702-style reliability gatekeeping; the Advisory Committee on Evidence Rules voted 8–1 to publish it on 2 May 2025 and public comment ran to 16 February 2026, but in early June 2026 the Standing Committee declined to recommend adoption and remanded it for revision and further study, together with the separate deepfake-authentication proposal for Rule 901(c). Machine-generated evidence is therefore handled under existing rules and judicial discretion, which is unsettled as of September 2026 and is exactly the condition in which a credible sponsoring expert does the most work.

What this article establishes

  • Proposed FRE 707 would have applied Rule 702-style reliability gatekeeping to machine-generated evidence offered without a human expert.
  • It stalled: published for comment on an 8–1 Advisory Committee vote in May 2025, then remanded for further study by the Standing Committee in early June 2026.
  • The deepfake-authentication proposal for Rule 901(c) has never been published for comment and was remanded alongside it.
  • With no dedicated rule, admissibility runs through existing authentication and expert rules and through the discretion of the individual judge.

What was proposed Federal Rule of Evidence 707 meant to do?

Proposed Federal Rule of Evidence 707 would have subjected machine-generated evidence offered without a sponsoring human expert to the same Rule 702-style reliability gatekeeping that expert testimony receives. The gap it addressed is straightforward: when an expert testifies, Rule 702 gives the court a mechanism to test reliability, and when a machine output is offered directly, it does not.

The practical target is the growing category of outputs that look like findings — algorithmic analyses, model-generated reconstructions, automated scoring — offered as evidence in their own right rather than as the basis of an expert’s opinion.

What happened to Rule 707?

It stalled and was returned to the Advisory Committee for revision and further study in June 2026. The sequence: the Advisory Committee on Evidence Rules voted 8–1 on 2 May 2025 to publish the proposal, with the Department of Justice dissenting; public comment ran from 15 August 2025 to 16 February 2026; at its 7 May 2026 meeting the Advisory Committee reported greater overall concerns arising from the comment file; and in early June 2026 the Standing Committee declined to recommend adoption.

A remand for further study is not a rejection of the underlying problem. It does mean that as of September 2026 there is no dedicated federal rule for machine-generated evidence and no published timetable for one.

Is there a rule for deepfakes?

No — the deepfake proposal is at committee-study stage and has never been published for public comment. The Advisory Committee has for roughly two years been considering a new Rule 901(c) that would allocate burdens when a party claims that proffered audio or video is an AI fabrication, and as of August 2026 it remained under study.

It was remanded alongside Rule 707 in June 2026, which is a signal worth noting: the committee is treating machine-generated evidence and machine-fabricated evidence as related problems rather than separate ones.

So how is machine-generated evidence actually handled now?

Through the existing rules and through the discretion of the individual judge, which in practice means authentication under Rule 901, the hearsay rules, and Rule 702 wherever a human expert is sponsoring the output. There is no uniform national answer, and outcomes vary by judge and by how the proponent frames what the machine produced.

That variability cuts both ways. A proponent with a credible expert prepared to explain the method has a materially easier path than one offering an output on its own, and an opponent has correspondingly more room to challenge an unsponsored output than the draft rule would eventually have required.

What does this mean for a litigator preparing a case in 2026?

Assume you may be asked to establish reliability without a rule telling you how, and build the record for it early. Practitioners should expect Rule 707-style reliability arguments to be raised on the strength of the proposal even though it was not adopted, and should check standing orders in each case, because judicial practice on AI has been moving faster than the rules.

The related development attorneys should track is judicial use itself. A 2026 survey of federal judges found more than 60% using at least one AI tool in chambers, mostly for legal research and document review, with roughly 25% formally permitting chambers AI use and about 20% banning it. California adopted the first comprehensive state judicial-branch AI rule, Rule 10.430, on 18 July 2025, effective 1 September 2025, requiring courts that permit generative AI to adopt a use policy by 15 December 2025.

What is genuinely unsettled here?

Nearly all of it, and that is the accurate answer rather than a hedge. The fate of Rule 707 after remand, the burden allocation for contested deepfakes under any eventual Rule 901(c), and judicial acceptance of generative-AI-assisted review protocols are all open as of September 2026.

What is not unsettled is the direction of travel: the questions courts are being asked to resolve increasingly turn on whether a method is reliable and who is prepared to stand behind it. The Institute’s AI Law & Rules area tracks these developments as they move.

For informational purposes only. Not legal advice and not ethics advice. Professional conduct rules are adopted state by state and diverge, and this record changes monthly. Anything here that reads as a holding should be checked against your own jurisdiction before it is relied on.

Related

The practice area

AI adoption conciergeorientation · not legal or ethics advice
Happy to. Tell me roughly how big the firm is and what it already pays for — Microsoft 365, Google Workspace, a practice-management system — because the honest answer to most AI questions at a firm your size starts with what you have already bought rather than what you should go and buy.