Anthropic co-founder rules out giving up oversight of Claude use

Anthropic co-founder Benjamin Mann said on October 14 that the company will not hand customers full control over how its most capable AI models are used, pointing to what he called the "dual use" character of the technology, Fortune reported.

Mann was responding to a Fortune question about whether Anthropic would store data locally and build sovereign AI offerings for governments and companies seeking tighter control of sensitive material. He said Anthropic wants customers to run demanding, sensitive work on its systems, but also wants to deploy the technology in a way that lets it identify and stop abuse.

In September, Anthropic launched its Enterprise Frontier Safeguards program, which includes zero data retention for qualifying customers alongside misuse detection. The company regularly publishes accounts of attempted misuse of Claude, including an alleged effort by a weapons cell in northern Yemen tied to ballistic missile work.

Mann spoke in Singapore at the opening of Anthropic's fifth Asia-Pacific office. More than 1,500 engineers at the state technology agency GovTech use Claude Code, according to the company, and DBS, GIC and Grab also work with the model.

OpenAI opened a Singapore office in late 2024 and has pledged 300 million Singapore dollars, or $234 million, to the local AI ecosystem. Mann criticized open-weight releases, saying such models cannot carry safeguards that hold up once published.

In June, U.S. export restrictions on Anthropic's Fable 5 and Mythos 5 models led the company to cut off access globally; the curbs were lifted within three weeks.

This story was first reported by Fortune.