Episode 023 Deepfakes and the Liar's Dividend

Has AI fundamentally changed the law of evidence, or has it just forced us to revisit principles that have always been there? That's the question at the center of this episode, and there's no better person to answer it than retired federal judge Paul Grimm — one of the people who has literally tried to write the new rule.

IN THIS EPISODE:

- Judge Paul Grimm's background: 25+ years on the federal bench in Maryland, now emeritus director of the Bolch Judicial Institute at Duke Law
- The multi-year federal rulemaking process (the Rules Enabling Act) and why it can't keep pace with generative AI
- The difference between "acknowledged" and "unacknowledged" AI-generated evidence, and why the distinction matters
- Proposed Rule 901(c) for authenticating AI-generated evidence — and why the Evidence Rules Advisory Committee has still declined to publish it after three years
- How the existing 51% authentication standard (Rule 901(b)(1)) creates a low bar for admitting evidence, real or fake
- The "liar's dividend": how the mere existence of deepfakes lets litigants dismiss real evidence as fabricated
- State of Washington v. Puloka (No. 21-1-04851-2 KNT, 2024): a court excludes AI-enhanced video evidence despite a legitimate clarification goal
- State of Florida v. Albisu (Broward County, 2025): a defense lawyer uses AI-generated VR goggles to let a judge see a self-defense claim from the client's point of view
- Rule 107, the new federal rule distinguishing "illustrative" AI-generated exhibits from actual evidence
- Why ABA Formal Ethics Opinion 512 already tells lawyers exactly how to avoid AI hallucination sanctions — and why lawyers keep getting sanctioned anyway

WE ALSO DISCUSS:

- The "witness washing" problem in facial-recognition photo lineups
- How deepfake detection expert Hany Farid's confidence has shifted as the technology improved
- Maura Grossman's deepfake-detection research at the University of Waterloo
- Judge Grimm's "wow moment" using AI to pressure-test a speech outline
- The origin of the term "deepfake" and how fast it entered the mainstream
- Why illustrative AI evidence — not fake evidence — may be the bigger near-term risk

KEY TAKEAWAY:

AI hasn't broken the rules of evidence. It's exposed how thin they always were. A 51% authentication standard and a jury instruction to "disregard" what they just saw and heard were never built for a world where anyone can manufacture convincing fake audio and video for free — and Judge Grimm's own proposed fix has been sitting in committee for three years.

For solo and small firm lawyers, the fix isn't waiting on Washington. Notice, disclosure, and pretrial motion practice around AI-generated evidence are tools you can use right now, rule change or not. Lawyers still dabbling with AI without a real verification habit are exposed either way — and the ones building real workflow discipline around it are the ones who'll actually benefit from what these tools can do.

MENTIONED IN THIS EPISODE:

- Judge Paul Grimm (Ret.), Bolch Judicial Institute, Duke Law
- Professor Maura Grossman, University of Waterloo
- Hany Farid, Dartmouth (formerly UC Berkeley)
- Federal Rules of Evidence 901(b)(1), 901(b)(5), 901(b)(9)
- Proposed Rule 901(c)
- Federal Rule of Evidence 403
- Federal Rule of Evidence 107 (illustrative evidence)
- State of Washington v. Puloka (No. 21-1-04851-2 KNT, 2024)
- State of Florida v. Albisu (Broward County, 2025)
- ABA Formal Ethics Opinion 512 (August 2024)
- "Learned Hand" (AI tool for courts)
- ChatGPT, Claude, Gemini
- The Rules Enabling Act

YouTube: https://youtube.com/@AIToolsforLawyers

Instagram: https://instagram.com/aitoolsforlawyers

[email protected]