Representatives from OpenAI and Meta Platforms told a federal judicial panel on Thursday that AI-generated deepfakes have become so advanced that verifying the authenticity of audio and video evidence in court will grow more difficult, Reuters reported.
The technology experts presented their concerns at a meeting in Boston of the US Judicial Conference's Advisory Committee on Evidence Rules — the body responsible for recommending changes to the federal rules that govern what evidence courts can admit. The committee has been working since 2023 on how to address the risks of fabricated AI-generated evidence in courtrooms. For more context on this story, see our ongoing artificial intelligence updates.
AI Labs Say Detection Has No Scientific Future
Shira Anderson, policy counsel at OpenAI, told the committee that AI images continue to improve in quality, and said she does not believe a scientific method to detect them will exist in the future — a striking admission from a company that builds the generative systems producing that content.
Devin Norder, director of product management at Meta, told the panel that media created by recent AI models cannot be distinguished by humans from authentic content, and that creating such media now requires minimal cost and time.
Both companies have developed provenance and verification tools for visual content, and new laws in California and the European Union require watermarking of AI-generated media. But Anderson warned the committee that each company uses its own system, meaning no universal method exists to reliably detect all watermarks or verify the authenticity of visual content across the ecosystem.
Courts Have Not Yet Felt the Wave
The committee's chair, US District Judge Jesse Furman of Manhattan, offered a counterweight to the alarm: while academic commentary predicts courts will deal with deepfakes, the issue has not yet appeared frequently in actual cases.
A survey conducted by the Federal Judicial Center in March underscored that gap. Only 15 out of 931 judges who responded had encountered a case where a party challenged evidence on the grounds that it was a deepfake.
That gap between prediction and practice is the committee's core dilemma. Under Federal Rules of Evidence 901 and 902, the existing standard for authenticating evidence is — as the committee's own report put it — a mild one: a party needs only to produce evidence sufficient to support a finding that the item is what the proponent claims it is. Deepfakes stress-test that standard, because the traditional pathways for authenticating recordings (witness testimony, system documentation) assume that forgery at scale is hard, expensive, and detectable.
The committee's own reporting has framed the problem with unusual bluntness. Its December 2025 report examined what happens when a video or audio offered at trial is a deepfake, concluding that it presents a direct challenge to the authenticity requirements under Rules 901 and 902 precisely because the existing standard of authenticity is a mild one. How — and whether — to harden that standard is the question the committee has been circling for three years, and it becomes more urgent if, as the lab representatives warned, detection science never catches up with generation science.
What Practitioners Can Do Now
Until the rules change, the practical burden falls on litigators and trial judges. Provenance tools — cryptographic content credentials embedded at creation time — are spreading through camera apps and creative software, though they only help for media that carries them. Cross-examination under oath of the person who captured a recording, discovery demands for original files and generation metadata, and closer scrutiny of unsourced clips circulating on social platforms are the current defenses. The advisory committee's message this week was effectively that these stopgaps will need a formal rules-level foundation, because the informal ones assume a detectability that OpenAI and Meta say no longer exists.
Why the Labs' Testimony Matters
It is notable that the companies whose products create the risk are the ones warning the judiciary about it. OpenAI and Meta have direct commercial and reputational interests in how the rules land: strict authentication requirements could impose friction on legitimate uses of AI media, while lax ones could make their tools vectors for courtroom fraud. Their message to the committee was effectively that provenance standards — watermarking, content credentials, chain-of-custody documentation — are the workable path, because detection alone will not scale.
The committee's three-year review is expected to inform whether the federal evidence rules need amendment for the AI era. Any change would move through the Judicial Conference's rulemaking process, including public comment, before going to Congress.
The Broader Regulatory Picture
The hearing lands amid a burst of deepfake-related policy activity. California and the EU have enacted watermarking mandates. Denmark has advanced legislation banning the sharing of AI deepfakes without consent. US courts have already seen AI-generated fabrications surface in filings, and judges have sanctioned lawyers for submitting fake citations generated by chatbots.
For now, the federal judiciary is choosing careful study over rapid rule changes. But the testimony from the companies building the technology suggests the industry itself does not expect detection tooling to keep pace — which shifts the burden toward provenance standards, procedural safeguards like live testimony requirements, and ultimately the judgment of judges and juries about what to believe.
---
Stay Ahead of AIGet the latest AI news, analysis, and breakthroughs — all in one place.
Read more AI news →
