AI Evidence Has No Special Rule Yet: Why FRE 707 Stalled and How to Handle AI and Deepfake Evidence Now
10/5/20264 min read
AI Evidence Has No Special Rule Yet: Why FRE 707 Stalled and How to Handle AI and Deepfake Evidence Now
If you've read that the federal courts will have a dedicated AI-evidence rule by December 2027, that timeline is no longer accurate. Proposed Federal Rule of Evidence 707 — the rule meant to govern machine-generated evidence — stalled in 2026. There is, at the moment, no special federal rule for AI-generated or deepfake evidence, and the one that was closest to adoption has been sent back to the drawing board.
For litigators, that's not a reason to wait. It means you'll be handling AI-generated output and deepfake challenges under the rules you already have, for the foreseeable future. Here's what actually happened, and how to work in the gap.
What happened to Rule 707
The Advisory Committee on Evidence Rules spent more than a year developing proposed Rule 707, titled "Machine-Generated Evidence." The timeline is worth knowing because it explains why the rule is now in limbo:
The Committee voted 8–1 in May 2025 to publish the draft for public comment, with the Department of Justice dissenting. The Standing Committee on Rules of Practice and Procedure approved publication in June 2025. Public comment ran from August 15, 2025 to February 16, 2026, with hearings in January 2026. Then the momentum reversed. At its May 7, 2026 meeting, the Advisory Committee declined to advance the rule in its current form, citing broad concerns raised in the comments. In June 2026, the Standing Committee decided not to recommend action, sending the proposal back for revision and further study. The Committee has scheduled a mini-conference with technical experts and practitioners at its fall meeting on October 15, 2026.
The practical upshot: the once-cited December 1, 2027 effective date is off track, and no adopted rule exists.
What Rule 707 would have done — and what it wouldn't
Understanding the proposal clarifies the gap it was supposed to fill. Rule 707 would have applied the reliability gatekeeping of Rule 702 — the standard for expert testimony — to machine-generated output offered without a supporting expert witness. The concern was straightforward: a lay witness might run a program and present its output while knowing little about whether the program is reliable. Rule 707 would have required that such output meet Rule 702's reliability requirements before a court admits it.
Two limits matter. First, the proposal applied only to evidence the proponent acknowledges is machine-generated. Second — and this is the point most coverage misses — Rule 707 was never the deepfake rule. It addressed the reliability of output a party admits a machine produced. It did nothing about authenticity: fabricated evidence passed off as genuine. A major reason the rule stalled was that it never clearly defined "machine-generated evidence," leaving its scope uncertain enough to invite early litigation.
The two problems, and why only one is even on a track
This is the distinction every litigator should internalize now, because the rulemakers have formally separated them:
Reliability — is the machine's output trustworthy enough to admit? That was Rule 707's territory, and it's back in revision.
Authenticity — is this video, image, or recording real, or an AI-generated fake? The Committee peeled this off into a separate proposal, Rule 901(c), to address claims that evidence is an AI deepfake. As of now, Rule 901(c) has not even been published for public comment.
So the reliability rule is paused, and the authenticity rule hasn't started its public process. Both problems are live in courtrooms; neither has a rule.
How to handle AI and deepfake evidence under the rules you already have
The existing Federal Rules were not written for AI, but they're what you have — and they're more capable than the gap suggests. A working approach:
Authentication (Rule 901). The proponent must show the item is what it's claimed to be. For digital media, that increasingly means more than "a witness recognizes it." Preserve and be ready to present metadata, source files, chain of custody, and the circumstances of capture. Rule 901(b)(9) — authenticating a result by describing the process or system that produced it — is a natural anchor for machine output.
Reliability (Rule 702). Where AI output amounts to an opinion or inference that would require expert testimony, treat it that way: be prepared to show it rests on sufficient data, reliable methods, and a reliable application of those methods. Don't let acknowledged AI output in through a side door simply because no rule yet names it.
Prejudice (Rule 403). Juries may over-trust anything that looks technological. Where AI-derived evidence carries a risk of misleading the jury that outweighs its probative value, 403 is available.
The "deepfake defense." Anticipate the mirror-image problem: opponents claiming genuine evidence is an AI fake to get it excluded. Courts are wary of letting a bare accusation of "deepfake" defeat otherwise authenticated evidence. Build your authentication record so a conclusory fabrication claim doesn't carry the day — and if you're the challenger, be prepared to offer more than suspicion.
Underneath all of this sits an attorney-competence point: you cannot authenticate, challenge, or cross-examine AI evidence you don't understand. The duty to grasp how these systems produce output — and where they fail — is now part of trial preparation, not optional background reading.
FAQ
Is Federal Rule of Evidence 707 in effect? No. It was published for comment, then sent back for revision in 2026. It has not been adopted, and the previously cited December 2027 date is no longer on track.
Does Rule 707 solve the deepfake problem? No. It addressed the reliability of acknowledged machine-generated output. The deepfake/authenticity question was split into a separate proposal, Rule 901(c), which has not yet been published for comment.
So how do courts handle AI evidence right now? Under existing rules — primarily authentication (901), expert-reliability (702), and prejudice (403). There is no dedicated AI-evidence rule in force.
Does this only affect federal court? The federal rulemaking is the clearest story, but state courts face the same gap under their own authentication and reliability rules. The practical burden on litigators is similar.
What should I watch next? The Advisory Committee's post-October 2026 output — a revised Rule 707 draft and any movement toward publishing Rule 901(c). Both remain proposals until they complete the full rulemaking path.
The rules are behind the technology, which means the burden of getting AI evidence admitted — or kept out — sits with the litigator, not the rulebook. If your team wants to be ready to authenticate, challenge, and cross-examine AI-generated and deepfake evidence under the rules as they stand today, that's exactly the kind of practical readiness we help attorneys build.