OpenAI, Meta caution US judiciary about risk of 'deepfake' evidence
META•OpenAI and Meta representatives warned a federal judicial panel that increasingly sophisticated AI-generated media will make it harder to authenticate audio and video evidence at trial. A survey found that 15 of 931 responding judges had encountered a case involving a challenge to evidence as a deepfake.
1. Rising authentication concerns
Tech experts at OpenAI and Meta Platforms told the US Judicial Conference's Advisory Committee on Evidence Rules that fabricated videos are becoming difficult for people and software to identify. OpenAI policy counsel Shira Anderson said she did not think there would be a scientific way to detect AI images going forward.
2. Limited cases reported so far
The Federal Judiciary Center said in March that 15 of 931 judges who responded to a survey had encountered a case in which a litigant challenged evidence because it was a deepfake. Judge Jesse Furman, who chairs the panel, said courts had not yet seen a lot of such cases, while experts warned that detecting fabricated media would become more challenging. Meta product management director Devin Norder said media from recent frontier models can be indistinguishable from authentic media to humans, and that creating it takes little time and money.
3. Detection tools have limits
OpenAI and Meta have sought to develop tools and services to verify visual content, while new laws in California and the European Union require watermarking aimed at detecting deepfakes. Anderson cautioned that there is no universal way to reliably detect all watermarks or verify visual content, and said she did not think detection would ever be bulletproof.
