Proposed Federal Rule of Evidence 707: what it means for photos and AI analysis

Courts are trying to decide how to treat evidence that a machine produced rather than a person. The most direct federal attempt is proposed Federal Rule of Evidence 707. It is not law. But the questions it asks are already the questions a careful judge or opposing lawyer will ask about any AI tool's output, including a photo analysis report.

What the proposed rule says

The version published for public comment in 2025 said, in short: when machine-generated evidence is offered without an expert witness, and it would be subject to Rule 702 if a person had given it as testimony, the court may admit it only if it meets Rule 702's reliability requirements. Output from basic scientific instruments was excluded.

Rule 702 is the rule for expert testimony. It asks whether:

• the output would help the court or jury;
• it is based on sufficient facts or data;
• it comes from reliable principles and methods; and
• those methods were reliably applied to this case.

The point of 707 is simple: a party should not be able to skip that scrutiny by handing the court a machine's answer instead of putting an expert on the stand.

Where it stands (as of October 2026)

• Public comments closed on February 16, 2026. They were split between support, support with changes, and opposition.
• At its May 2026 meeting, the federal Advisory Committee on Evidence Rules did not advance the published version. It kept working on a revised draft, which would generally expect an expert to explain how the AI system reliably produced its output.
• The committee planned to hear from technical experts and practitioners at its fall 2026 meeting. A revised rule would likely need to be published for comment again, so any federal rule is years away.

Check the federal judiciary's rules committee pages for the latest official status. Some states are working on their own approaches.

Why it matters even though it isn't law

Judges already have tools to question unreliable machine output: Rule 702 when an expert relies on it, Rule 901 on authentication, and Rule 403 on unfair prejudice. Proposed 707 mostly gives a name to a standard that careful courts apply anyway. If you plan to rely on any software's conclusion about a photo, expect to answer the 702 questions.

What that means for a photo analysis report

A single "AI probability: 87%" score is hard to defend under a reliability test. Nobody can say what facts it rests on, how the method was validated, or whether it applies to this particular file. A report holds up better when it can show:

1. The facts it used. Which signatures, metadata fields and file structures it read, listed so someone else can look at them.
2. A known method. Strong evidence, such as a cryptographic signature or an exact hash match, is kept separate from weak hints, such as metadata that anyone can edit.
3. Its limits, stated up front. What the evidence cannot establish, and when the honest answer is "unknown".
4. Reproducibility. The same file gives the same answer, and the file's hash ties the report to that exact file.
5. Testing. How the tool behaves on files whose history is known, including untouched camera originals, not just edited ones.

How PIT is built with this in mind

PIT does not give a real-or-fake score. It answers four questions separately: how the image was created, whether it was processed or altered, where it came from (origin), and how strongly that evidence is tied to this exact file (integrity). Each answer lists the evidence behind it and its limits. Strong evidence, such as valid Content Credentials from a recognized signer, is kept apart from weak evidence, such as EXIF fields. When the evidence can't support a conclusion, PIT says "unknown".

Reports are digitally signed and include the photo's SHA-256 hash, so anyone can later check that a report is unchanged and refers to the same file.

None of that decides admissibility; the court does. And a PIT report does not replace a qualified forensic examiner when a case needs expert testimony. It gives you, your expert, or your opponent a clear, checkable record of what the file itself shows.

FAQ

Is Rule 707 in effect?

No. As of October 2026 it is a proposal under revision, not a rule.

Would 707 cover deepfakes?

Not directly. 707 is about the reliability of machine output offered as evidence. The question of whether an exhibit itself is a deepfake is a matter of authentication under Rule 901, which the committee has studied separately. See our guide to authenticating photos under Rule 901.

Can a PIT report be used in court?

That is for the court to decide in each case. PIT is designed so its conclusions can be traced to evidence and its limits are stated, which is what a reliability inquiry looks for.

This guide explains evidence rules in general terms. It is not legal advice, and rules and local practice vary by court. Talk to a lawyer about your own case.