Anthropic CEO denounces OpenAI Pentagon pact as “safety theater” as both firms re‑enter

Anthropic has re‑opened talks with the Defense Department after the Pentagon labeled the startup a “supply‑chain risk” on Friday, according to CNBC. The label followed Anthropic’s refusal to let the military use its models for mass domestic surveillance or fully autonomous lethal weapons. The move comes as OpenAI announced a separate contract to place its AI on classified systems, a deal Sam Altman later described as “sloppy” in a Bloomberg interview.

In a 1,600‑word memo leaked to staff, Anthropic CEO Dario Amodei called OpenAI’s Pentagon agreement “80 % safety theater,” accusing Altman of trying to “undercut our public support” and give the Trump administration a lever to punish Anthropic, The‑Decoder reported. Amodei’s internal letter, sent to more than 2,000 employees, framed the dispute as a broader clash over political loyalty rather than a simple contract negotiation.

Investors have reacted swiftly. The Decoder noted that capital providers are pressing both firms for de‑escalation, while a major tech industry group has publicly backed Anthropic’s stance. Amodei is reportedly making a “last‑ditch” appeal directly to the Under Secretary of Defense for Research and Engineering, hoping to salvage a deal that would keep Anthropic in the defense supply chain, per The‑Decoder.

OpenAI, meanwhile, is racing to finalize its own agreement. Bloomberg said Altman acknowledged the Pentagon deal was “sloppy” and that the company is working to tighten the terms. The contract would embed OpenAI’s models in classified environments, positioning the firm as the primary AI supplier for the department.

The parallel negotiations underscore a widening rift in the AI sector. As Anthropic pushes back on military use cases it deems unsafe, OpenAI moves forward with a government partnership that could set a precedent for future defense contracts. Both companies face heightened scrutiny from regulators, investors, and a public wary of AI’s role in national security.