OpenAI, Anthropic lobby on antitrust cover for AI security sharing

The filings put a concrete paper trail behind an awkward tension in AI policy. Labs want to compare notes on security risks — Anthropic's form names the Collaboration on Adversarial Threats and Security Risks Act — yet doing so can look like the kind of coordination antitrust law is designed to catch. That is why the ask is for a narrow protection rather than a broad exemption.

The timing connects to last month's talk of a coordinated AI development "slowdown" pitched by AI leaders. OpenAI's Chris Lehane has argued an antitrust exemption specifically for model pacing would not be necessary, and Trump administration officials criticized AI leaders for seeking such a carve-out, saying they could decide to slow down among themselves without government blessing. The narrower security-sharing framing, rather than pacing, appears to be where the lobbying paperwork sits. Whether any of it becomes law may depend on the defense policy vehicle Semafor previously reported could carry the collaboration bill.