OpenAI and Meta Warn of Deepfake Threats to Legal Evidence
Representatives from OpenAI and Meta warned a U.S. judicial advisory committee on October 15 that rapidly improving AI-generated deepfakes pose a severe threat to the reliability of evidence in legal proceedings. They noted that humans can no longer distinguish real media from outputs generated by frontier AI models, and creation costs are extremely low. This warning highlights a critical societal and legal challenge as generative AI tools become more accessible and realistic. The erosion of verifiable evidence could severely undermine the integrity of judicial systems worldwide if courts cannot effectively detect fabricated media. While major AI companies are adding visible or invisible watermarks to AI-generated media, there is currently no universal method to detect these watermarks. However, deepfakes remain relatively rare in current courtroom practice, with a survey of 931 U.S. judges showing only 15 had encountered cases where evidence credibility was questioned due to suspected deepfakes.
## BACKGROUND
AI watermarking is a proactive technique that embeds imperceptible signatures or metadata into AI-generated outputs to enable origin verification. Deepfakes refer to hyper-realistic manipulated digital media—including images, audio, and video—produced using deep learning algorithms that can convincingly mimic real people or events.