Federal Rules Committee Still Reviewing AI Expert Witnesses

After more than three years of review, the federal Advisory Committee on Evidence Rules has decided not to advance Proposed Rule 707 artificial intelligence (AI) evidence, opting instead to gather further input.

The Committee approved Proposed Rule 707 for public comment back in May 2025. Then in May 2026, the Committee issued a memorandum that it had decided to gather further input, including a Fall 2026 mini-conference regarding draft Rule 707.

The American Bar Association has came out emphatically against Rule 707, calling it redundant (See ABA Redundancy Proposal Rule 707.) The ABA argued that the rule is unnecessary because existing expert testimony rules already handle sophisticated computer analysis.

Current Status of Proposed Rule 707

• The Proposal: Drafted to govern AI-generated and machine-produced information offered at trial without a human expert.
• The Framework: Designed to apply Daubert-style reliability standards—similar to expert testimony under Rule 702—to complex machine outputs. 
• The Pushback: Received dozens of critical public comments calling the initial draft overbroad, premature, or likely to trigger excessive satellite litigation.
• Next Steps: The Committee paused the current draft and scheduled a mini-conference with technical experts and practitioners for its fall meeting on October 15, 2026.

The Proposed Rule 707 is modeled directly after the four reliability criteria used for human expert witnesses under Rule 702. The rule was drafted to create a pretrial judicial gatekeeping mechanism. This stops parties from bypassing strict reliability checks when introducing complex data.

Under the initial draft of Rule 707, a court can only admit machine-generated evidence if the proponent demonstrates that:

• Sufficient Data: The output is built on an adequate foundation of underlying facts or data.
• Reliable Principles: The technology uses scientifically valid and dependable methods.
• Dependable Processes: The software or hardware applies those principles accurately.
• Accurate Application: The specific output in question correctly reflects the system's reliable application to the facts of the case.

Groups like the American Association for Justice (AAJ) have warned that the term "machine-generated" sweeps in routine evidence like GPS metadata, surveillance footage, and electronic health records. Others have commented on "vague exemptions," such as the "simple scientific instruments" exception—meant to exempt basic tools like digital scales. This was widely criticized as too ambiguous, threatening to spark massive satellite litigation.

Interestingly, while the American Bar Association (ABA) has called the Rule redundant, the New York City Bar Association has argued that traditional authentication rules fail to check for hidden algorithmic errors, algorithmic bias, or systemic errors.

While the Committee dithers, expert testimony using AI is surging, with no rule and no guardrails. Then again, it appears expert witnesses need not worry about being quickly replaced by AI — unlike many professionals.

[Note: This article was written in part by AI Claude.]