OpenAI, Meta caution US judiciary about risk of 'deepfake' evidence
By Nate Raymond
Oct 15 (Reuters) - Tech experts at OpenAI and Meta Platforms META.O warned a federal judicial panel on Thursday that AI-generated "deepfakes" are becoming so sophisticated that it will become increasingly challenging to determine the authenticity of audio and video evidence presented at trial.
The companies' representatives and other experts shared their views on the evolving technology during a meeting in Boston of the US Judicial Conference's Advisory Committee on Evidence Rules, which for three years has grappled with how to police the risks of fabricated AI-generated evidence in court.
In that period alone, rapid advancements in generative AI have reached the point where fabricated videos have become difficult for both humans and software to spot, experts advised at the meeting.
"AI images are only getting better, and I do not think that there will be a scientific way going forward to detect them," OpenAI policy counsel Shira Anderson told the committee.
The panel has struggled since 2023 with how to address authentication issues related to deepfakes and ensure that evidence known to be generated by AI meets the same reliability standards as evidence from a human expert witness.
The judiciary has sought to get ahead of the issue despite a lack of evidence that it is a widespread problem. The Federal Judiciary Center, the judiciary's research arm, in March said that out of 931 judges who responded to a survey, only 15 had encountered a case in which a litigant challenged the introduction of evidence because it was a deepfake.
"There's a lot of academic commentary and other commentary predicting that courts will be dealing with deepfakes, but we haven't actually seen a whole lot of that yet," said US District Judge Jesse Furman of Manhattan, who chairs the panel.
Experts from industry, academia and government warned, though, that the issue is far from purely hypothetical. They said the courts' ability to act as gatekeepers against fabricated AI-generated evidence is only going to become more challenging.
"Media generated from frontier models, at least in the last few generations, are indistinguishable by humans from authentic or AI-generated media," said Devin Norder, director of product management at Meta. "And the cost and the time it takes to create this media really are quite trivial at this point."
Major companies including OpenAI and Meta have sought to adopt tools and services aimed at verifying the validity of visual content, and new laws in California and the European Union require "watermarking" aimed at detecting deepfakes.
But Anderson warned that because each company has its own system, there's no universal means to reliably detect all watermarks or verify the authenticity of visual content.
"My guess is that this absolutely will continue to get better," she said. "I do not think it'll ever be bulletproof."
Read more:
US judicial panel delays action on AI-generated evidence, deep fakes
Lawyers doubtful about US judiciary's draft rule for AI-generated evidence
US judicial panel advances proposal to regulate AI-generated evidence
US judicial panel to develop rules to address AI-produced evidence
US judicial panel wrestles with how to police AI-generated evidence
US appellate judge calls bans on AI use by lawyers 'misplaced'
US Supreme Court's Roberts urges 'caution' as AI reshapes legal field