OpenAI and Meta have warned US judges that increasingly convincing artificial intelligence-generated recordings threaten to complicate one of the courtroom's most fundamental tasks: determining whether evidence is genuine.

The warning came during an October 15 meeting of the US Judicial Conference's Advisory Committee on Evidence Rules in Boston, where judges are considering whether existing safeguards can withstand sophisticated digital forgeries.

OpenAI policy counsel Shira Anderson questioned whether a reliable scientific method for detecting AI-generated imagery would ever be developed.

Meta product management director Devin Norder told the committee that media produced by the latest advanced AI models could be indistinguishable from authentic recordings to human observers. Norder also warned that producing such material now required remarkably little time or expense.

No universally reliable method for detecting deepfakes has emerged. Although companies are developing tools to identify synthetic material, Anderson warned that differing watermarking systems make it difficult to establish whether digital content is authentic.

For now, documented courtroom challenges remain uncommon. A Federal Judicial Center survey released in March found that just 15 of 931 responding judges had encountered a case in which evidence was challenged as a deepfake. The figure measures objections raised, not the number of fabrications potentially submitted.

The committee has drafted a proposed Rule 901(c). Under the proposal, anyone alleging that evidence is fabricated would first need to provide sufficient grounds for that claim. The party introducing the evidence would then have to establish that it was more likely than not authentic.

The rule has not been approved. US District Judge Valerie Caproni urged caution, arguing that courts already have laws against fraudulent evidence.

Committee chairman Judge Jesse Furman has appointed a subcommittee to examine the issue further, leaving judges to rely on existing evidence rules as the technology advances.