Technical experts from OpenAI and Meta Platforms warned on October 15, 2026, Thursday, at a meeting in Boston of the Advisory Committee on Evidence Rules of the U.S. Judicial Conference: AI-generated "deepfake" materials are becoming so sophisticated that establishing the authenticity of audio and video evidence presented in court is getting harder. The committee has spent three years discussing how to manage the risk of fake AI-generated evidence in courts — the October 15 meeting was the latest stage of that discussion. Reuters covered the details of the meeting.
The Detection Threshold Is Narrowing
The experts told the committee that rapid advances in generative AI in recent years have taken fake videos to a level where they are hard to spot for both people and software tools. In their assessment, today's synthetic video can look authentic not only to an ordinary viewer but also to specialized software. This means the task of determining whether a video or audio recording submitted as court evidence is genuine or AI-generated is becoming ever harder.
Meta's Director of Product Management Devin Norder noted that media produced by the last few generations of frontier models is indistinguishable from the real thing to the human eye, while the cost and time needed to create such content are now negligible. In his words, frontier models keep improving from generation to generation, and telling synthetic media apart from the original has become nearly impossible.
Anderson spoke before the committee as an OpenAI policy advisor; in her assessment, there is little hope that a scientifically grounded method of exposing AI imagery will emerge as its quality keeps improving.
"AI images will only get better, and I don't think there will be a scientific way to detect them in the future," said OpenAI policy advisor Shira Anderson at the committee meeting.
The Committee Has Been Seeking a Solution for Three Years
The Advisory Committee on Evidence Rules is an advisory body to the U.S. Judicial Conference that helps shape the rules of evidence in federal courts. Over three years of discussions, its members have been working on how to manage the risk of fake AI evidence entering court proceedings — and technology has advanced significantly in that time.
Since 2023 the committee has been considering ways to address deepfake-related authentication problems. One of the central questions in the discussion is whether evidence known to be AI-generated must meet the same reliability criteria as the testimony of a human expert witness. The event was also reported by Yahoo Finance Canada on October 15, 2026. The Yahoo Finance Canada report also confirms that the meeting took place on October 15, 2026.
The deepfake problem is growing beyond the courtroom as well — aishunos.uz previously wrote about deepfake and disinformation.
Watermarking Laws Passed, but No Unified Standard Exists
OpenAI and Meta have already introduced tools and services aimed at verifying the authenticity of visual content. New laws adopted in California and the European Union require "watermarking" — embedding hidden digital marks in content — to detect deepfakes.
But as Anderson told the committee, because each company has its own system, there is currently no universal way to reliably detect all watermarks or to verify the authenticity of visual content. As she explained, each major company uses its own watermarking system — as a result, a mark placed by one company cannot be reliably read by another system. That is why, for now, there is no way to authenticate visual content with a single method.
The Judges Themselves Have Rarely Encountered It
The Federal Judicial Center — the research arm of the court system — in March 2026 published the results of a survey of 931 judges. Only 15 of them said they had handled a case in which a party objected to admitting evidence, arguing it was a deepfake. The survey results show that evidence objections over deepfakes remain rare in court practice.
"Many academic and other commentaries predict that courts will deal with deepfakes, but in practice we have not yet seen many such cases," said committee chairman Jesse Furman, a Manhattan federal district judge. He noted that while academic and public discussions contain many predictions about courts colliding with deepfakes, in practice the number of such cases is still small.
