Authenticating AI Evidence

Short Answer

Texas and federal authentication rules require evidence sufficient to support a finding that an item is what its proponent claims (Rules 901). Expert reliability rules can apply to analysis used to support the evidence (Rules 702). The federal committee withdrew the version of proposed Rule 707 released for comment and returned a modified proposal for further study. Its October 15, 2026 agenda is a future committee discussion, not an operative evidence rule. Reliability of an AI analysis and a claim that genuine evidence is a deepfake are different questions; both need original records and a clear account of the process.

Which Laws Apply

Texas AI-specific: none.

Generally applicable Texas law: Tex. R. Evid. 702, 705 and 901; Texas reliability case law for experts.

Federal: Fed. R. Evid. 702 (amended 2023), 901; proposed Fed. R. Evid. 707.

Two Kinds of AI Evidence

AI-generated evidence is output produced by a model, such as a risk score, a reconstruction or an analysis of data. The question is reliability: what data went in, how the model was validated, and whether its error rate is known. AI-altered evidence is a recording or image that has been changed or is alleged to have been fabricated. The question is authenticity: whether the item is what the proponent says it is.

Proposed Federal Rule 707

The version of proposed Rule 707 published for comment would have addressed certain machine-generated evidence without an accompanying expert. At its May 7, 2026 meeting the committee withdrew that version for further study. Its September 23 agenda book schedules discussion of modified text and deepfake authentication for October 15. The meeting had not occurred at this page’s check date. Existing rules remain the source of admissibility requirements.

Deepfake Challenges

When evidence is challenged as altered or synthetic, preserve the original file, metadata, provenance records and every processing step. A witness with knowledge can address capture and custody; technical testimony may be needed to explain an enhancement or model inference. Retain the source version alongside any enhanced version. Content provenance can support a factual showing but does not itself guarantee authenticity or admissibility.

Practical Steps

For AI-generated analyses, disclose the system, inputs, validation and error rates, usually through an expert.

For recordings and images, preserve originals and metadata from the moment of collection.

When challenging an exhibit, ask for native files and the processing history, not only the displayed version.

What Is Unsettled

Whether a federal Rule 707 will be adopted and whether Texas will follow; how much proof a party must offer to shift the burden when it claims evidence is a deepfake.