OpenAI and Meta warn judges deepfakes evade detection

OpenAI and Meta Platforms told a federal judicial panel on Thursday that AI-generated deepfakes have grown so advanced that courts will struggle to tell whether audio and video evidence is authentic, and the panel's chair appointed a subcommittee to study the problem, Reuters reported.

The representatives spoke at a meeting in Boston on October 15 of the Advisory Committee on Evidence Rules, which advises the US Judicial Conference on the Federal Rules of Evidence. The committee convened a panel of outside experts on deepfakes that morning, according to its agenda book for the session.

Shira Anderson, policy counsel at OpenAI, told the committee that AI images keep improving. "AI images are only getting better, and I do not think that there will be a scientific way going forward to detect them," she said, according to Reuters.

Devin Norder, director of product management at Meta, said media produced by frontier artificial intelligence models has become impossible for people to distinguish from authentic footage. "The cost and the time it takes to create this media really are quite trivial at this point," he said, according to Reuters.

The committee has worked on the authenticity problem since 2023 and has taken it up at seven meetings, its agenda book states. It has prepared but not approved a new Rule 901(c), which would require a party challenging an item as a deepfake to offer evidence supporting that claim, after which the party offering the evidence would have to show the item is more likely than not authentic.

Nothing was decided at the Boston meeting, and no amendment to Rule 901 was proposed. The committee declined to advance a deepfake amendment at its spring 2026 meeting, and in May 2025 it reasoned that "very few examples exist of courts having to address the possibility of deepfakes," according to a meeting report Legaltech News cited on May 7, 2026.

The hesitation rests on thin case volume. A Federal Judicial Center survey in March found that only 15 of 931 responding judges had seen a case in which a litigant challenged evidence as a deepfake, Legaltech News reported. The same survey found a little over half of the responding judges believed the rules for authenticating evidence should be changed.

Judge Jesse Furman, the US district judge for the Southern District of New York who chairs the committee, said the predicted wave has not yet arrived in court. "There's a lot of academic commentary," he said, "but we haven't actually seen a whole lot of that yet." He appointed a subcommittee to examine the issue, according to Reuters.

Both companies build tools meant to verify visual content. Anderson cautioned that each company uses its own system, so no single method reliably detects every watermark. New laws in California and the European Union require watermarking to flag synthetic media, Reuters reported.

For creators, the hearing tracks a problem audiences already argue about from the other end: when a synthetic clip of a real person circulates, the dispute now lands on standards written for photocopied documents. If courts adopt a burden-shifting test, the clip's sponsor rather than its accuser would carry the heavier proof. No federal rule specific to AI output or deepfakes is in effect or pending final approval.

The next step is the subcommittee's review of whether authentication rules need to change before the committee decides whether to advance any amendment.