Anthropic CEO memo attacks OpenAI over Pentagon deal

Anthropic CEO Dario Amodei published an internal memo criticizing OpenAI’s defense contract and the broader approach to AI safety inside the industry.

In the memo, Amodei calls the Pentagon partnership “safety theater” and argues that monitoring systems offered to AI companies are limited in practice. He claims jailbreaks still bypass safeguards and says monitoring works only in a small share of cases.

The memo also describes discussions with Palantir during negotiations. According to Amodei, Palantir proposed tools to detect violations inside AI systems, which he framed as monitoring rather than preventing misuse.

Amodei said Anthropic refused contract terms that included “analysis of bulk acquired data.” The Pentagon reportedly offered to accept Anthropic’s conditions if that phrase was removed, but the company declined.

Subscribe on @venture