AI Regulation Tracker / Courts and litigation

A Proposed Federal Rule Would Put AI-Generated Evidence Through the Same Reliability Gate as Expert Testimony

Proposed Federal Rule of Evidence 707 is still a draft moving through the rulemaking process, not law. If it advances, machine-generated evidence offered without a live expert would have to clear the Rule 702 reliability standard before a jury sees it.

By Anthony Guerriero · The Leveraged Years Regulatory Research · Published July 21, 2026 · Last updated July 21, 2026

The short version

Proposed Federal Rule of Evidence 707 is a draft rule that would treat machine-generated evidence, including AI output, the way courts treat expert testimony. Under the proposed text, where such evidence is offered without a sponsoring expert witness and would be subject to Rule 702 if a person testified to it, the court must find that it satisfies the Rule 702 reliability requirements before admitting it. It carves out the output of basic scientific instruments. Important status point: this is not law. The Advisory Committee on Evidence Rules published Rule 707 for public comment in August 2025, the comment period closed February 16, 2026, and the committee took the proposal back up at its May 7, 2026 meeting. Even if it keeps advancing, it still has to clear the Standing Committee, the Judicial Conference, the Supreme Court, and a Congress review window, so the earliest realistic effective date is late 2027. For litigators, the signal is clear now: build a reliability record for any AI-derived evidence you plan to offer.

Regulatory Briefing

Instrument: Proposed Federal Rule of Evidence 707 (machine-generated evidence), published for public comment
Authority: Advisory Committee on Evidence Rules, Judicial Conference of the United States; overseen by the Committee on Rules of Practice and Procedure
Jurisdiction: United States federal courts (would apply nationwide if adopted)
Bindingness: Non-binding at this stage. It is a proposed rule in the middle of the federal rulemaking process, not an enacted rule. It changes no current admissibility standard.
What Would Change: Machine-generated evidence offered without an expert witness would have to satisfy Rule 702 reliability requirements before admission, closing a gap where AI output could otherwise sidestep expert scrutiny.
Who This Affects: Trial attorneys, litigators, expert-retention teams, and any party planning to offer AI or algorithmic output as evidence in federal court.
Current Status: Public comment closed February 16, 2026; Advisory Committee reviewed comments at its May 7, 2026 meeting. Still subject to Standing Committee, Judicial Conference, Supreme Court, and Congress before it could take effect, earliest around December 2027.
Primary Source: U.S. Courts, Advisory Committee on Rules of Evidence, uscourts.gov
Last Verified: July 21, 2026

There has been a hole in the rules of evidence, and Rule 707 is the proposed patch. When a human expert takes the stand, their opinion has to clear Rule 702, the reliability gate that came out of the Daubert line of cases. The proponent has to show the testimony rests on sufficient facts, uses reliable methods, and applies those methods reliably to the case. But when a machine produces an expert-like conclusion and a party offers that output directly, without putting a human expert behind it, courts have not had a clean rule requiring the same scrutiny. As AI tools generate more of the analysis that used to come from a testifying expert, that gap started to matter.

The proposed text is short and pointed. In the version published for comment, Rule 707 provides that "where machine-generated evidence is offered without an expert witness and would be subject to Rule 702 if testified to by a witness, the court must find that the evidence satisfies the requirements of Rule 702(a)-(d)." It then adds a carve-out: "this rule does not apply to the output of basic scientific instruments." In plain terms, if a machine is doing what an expert would do, its output has to earn admission the same way an expert would. A thermometer or a basic measuring device does not.

What would actually change if Rule 707 is adopted?

The practical effect is that you could not launder an unreliable analysis through software to escape expert scrutiny. If a party wants to put in a predictive model's conclusion, an AI-generated reconstruction, or an algorithm's inference, and there is no expert on the stand to defend it, the court would have to run that output through the Rule 702 factors before it goes to the jury. That means a record on the underlying data, the methodology, validation, and how it was applied to the facts of this case. It does not ban AI evidence. It refuses to let AI evidence in on easier terms than the human expert it is standing in for.

Read the status precisely: this is a draft, not a rule yet

I want to be careful here because it is easy to talk about proposed rules as if they were already in force. Rule 707 is not law. It is a proposal working through the multi-step federal rulemaking process. The Advisory Committee on Evidence Rules voted to publish it for public comment, the comment period ran and closed on February 16, 2026, and the committee took the comments back up at its meeting on May 7, 2026. Publishing a rule for comment is explicitly not a promise that it will be enacted, and the committee said as much. From here it would still need to clear the Standing Committee, then the Judicial Conference, then the Supreme Court, and then survive a review window in Congress. That whole pipeline means the earliest a rule like this typically takes effect is December of the following year, so late 2027 at the soonest, and the text can still change or die along the way.

Why litigators should not wait for it to become final

Even as a draft, Rule 707 tells you where courts are heading, and the smart move is to get ahead of it rather than react to it. Judges are already wrestling with AI-derived evidence under existing rules, and the reasoning behind Rule 707, that machine output standing in for an expert should meet expert-level reliability, is the direction the bench is leaning regardless of whether this specific rule passes. If you plan to offer AI or algorithmic output in a federal case, build the reliability record now. Document the tool, its training and validation data, its error rates, and how it was applied to your facts. Decide whether you actually need a testifying expert to sponsor it, because that path already has a well-worn admissibility route. And on the other side, if your opponent is offering AI-generated evidence, this is your roadmap for the challenge: demand the reliability showing that Rule 707 would formalize. The rule is not binding yet, but the professional habit it points to is worth adopting today.

Source File

Primary: U.S. Courts, Committee on Rules of Practice and Procedure and Advisory Committee on Evidence Rules, uscourts.gov
Instrument: Proposed Federal Rule of Evidence 707, machine-generated evidence, published for public comment (August 2025), comment closed February 16, 2026
Key text: Where machine-generated evidence is offered without an expert witness and would be subject to Rule 702 if testified to by a witness, the court must find that the evidence satisfies the requirements of Rule 702(a)-(d); the rule does not apply to the output of basic scientific instruments.
How to verify: Open the Advisory Committee on Evidence Rules materials to confirm the proposed Rule 707 text, the August 2025 publication for comment, the February 16, 2026 comment deadline, and the May 7, 2026 committee meeting, and to confirm the rule has not yet been enacted.

Questions professionals are asking

Is Rule 707 in effect now?

No. It is a proposed rule in the federal rulemaking process. It was published for public comment in August 2025, comment closed February 16, 2026, and the Advisory Committee reviewed comments at its May 7, 2026 meeting. It still needs the Standing Committee, Judicial Conference, Supreme Court, and a Congress window before it could take effect, so late 2027 at the earliest.

What does the proposed rule actually require?

Where machine-generated evidence is offered without an expert witness and would be subject to Rule 702 if a person testified to it, the court must find that the evidence satisfies the Rule 702(a)-(d) reliability requirements before admitting it. The rule does not apply to the output of basic scientific instruments.

Does it ban AI-generated evidence?

No. It would require AI or machine output that stands in for an expert to meet the same reliability standard an expert would, rather than getting in on easier terms. AI evidence can still be admitted if the proponent makes the Rule 702 showing on data, methodology, and application to the case.

How is this different from just calling an expert?

If a testifying expert sponsors the analysis, the existing Rule 702 route already applies. Rule 707 targets the gap where a party offers machine output directly, with no expert on the stand, and asks the court to apply the same reliability gate so the absence of a human witness does not lower the bar.

What should trial attorneys do while it is still a draft?

Build the reliability record now for any AI-derived evidence you intend to offer: document the tool, its training and validation data, error rates, and how it was applied to your facts, and consider whether a sponsoring expert is the cleaner path. If an opponent offers AI evidence, use the Rule 707 logic to demand that reliability showing.

RELATED BRIEFINGS

Browse the full AI Regulation News tracker

Informational analysis for working professionals, not legal advice. Confirm how the Federal Rules of Evidence and any proposed amendment apply to your matter with qualified counsel.