Before there is an AI evidence rule, there is an evidence question.
Federal rulemakers are revisiting a possible Rule 707 for certain machine-generated evidence offered without an accompanying expert. The October 15 Evidence Committee agenda lists separate sessions on deepfakes and authenticity, and on artificial intelligence and proposed Rule 707.
The distinction matters. A deepfake dispute may begin with authenticity. The Rule 707 project addresses a different concern: potentially unreliable AI or machine output offered as evidence without an expert. The two questions can overlap, but they are not the same question.
The agenda also makes clear that Rule 707 is not a new federal evidence rule. The Committee received testimony at two hearings and more than 70 written comments on the public-comment version. At its last meeting, it withdrew that version and submitted a modified proposal for further study. The October 15 agenda includes an expert panel to discuss potentially unreliable AI offered as evidence and suggestions about a proposed rule.
That is a useful moment for caution. A new federal evidence rule is not in force. It does not change Tennessee evidence law. But the underlying proof questions are already here.
The practical lesson is not to wait for a new rule before doing the old work of laying a foundation.
Four foundation questions for AI-related evidence
- First, identify the item. Is the exhibit an original recording, an altered recording, a synthetic creation, an AI-assisted summary, or a system output? A label such as “AI-generated” may be too broad to answer the real question.
- Second, preserve the available lineage. Keep the original file when possible. Preserve available metadata, system records, version information, prompts, inputs, or other material that may show how the item was created or changed. A clean screenshot may be useful for display, but it is not always the whole record.
- Third, name the human witness or proof path. Someone should be able to explain the source of the item, the process that produced it, and the limits of what the item can establish. That may be a fact witness, custodian, system operator, expert, or another qualified source of proof. The answer will depend on the item and the case.
- Fourth, do not stop at authenticity. Even a well-identified item may leave other familiar evidence questions. The point of a foundation is not to end the inquiry. It is to give the court and the parties a reliable starting place for the inquiry that follows.
The current federal rulemaking record makes this plain. The October 15 Evidence Committee agenda places deepfake authenticity and proposed Rule 707 in separate tracks. The U.S. Courts’ public suggestion page also lists 2026 Rule 707 and Rule 901 submissions as pending consideration. No one should guess what the Committee will do next.
The safer conclusion is smaller and more durable: AI-related evidence needs a record that can be tested. The responsible question is not merely, “Was AI involved?” It is, “What is this item, how was it made, and what proof supports its use here?”
The question for this week
If an AI-related item were offered in court tomorrow, could the party offering it show its source, its process, the person who can explain it, and the limits that still matter?