OpenAI and Meta Warn US Federal Judiciary That AI Deepfakes Will Defy Scientific Detection

OpenAI and Meta $META representatives warned the US Judicial Conference's Advisory Committee on Evidence Rules in Boston on Thursday that AI-generated audio and video are becoming too sophisticated for humans or software to reliably authenticate at trial. A March survey by the Federal Judiciary Center found that out of 931 responding judges, only 15 had encountered a case where a litigant challenged evidence as a deepfake. OpenAI policy counsel Shira Anderson told the panel that because each company uses its own watermarking system under California and EU laws, there is no universal, bulletproof method to verify visual content. Meta product management director Devin Norder noted that media from recent frontier models is indistinguishable by humans from authentic media, with trivial creation time and cost. US District Judge Jesse Furman, who chairs the panel that has studied the issue since 2023, appointed a subcommittee on Thursday to examine the problem further as Judge Valerie Caproni cautioned against rushing an AI-specific rule. "AI images are only getting better, and I do not think that there will be a scientific way going forward to detect them." — Shira Anderson, OpenAI Policy Counsel.