On a Friday in late February, Dario Amodei sat contemplating the fate of his five-year-old company. It was the end of a surreal week for Anthropic, even by the standards of an extraordinary few months for the AI start-up that was speeding ahead of its rivals to develop cutting-edge systems. A bitter public clash had erupted with the US Department of Defense over a principle at the core of the company’s founding: the safe deployment of powerful AI technologies.
The DoD had asked Anthropic to allow its AI system Claude to be made available for “all lawful purposes”. Anthropic had asked for two exceptions: mass surveillance of Americans and lethal autonomous warfare. From there, the negotiations descended into mud-slinging.
So far that day, President Donald Trump had posted that every federal agency in his government should immediately cease working with Anthropic, describing Amodei and his colleagues as “woke . . . leftwing nut jobs”. Defense Secretary Pete Hegseth followed up with a post saying Anthropic would be designated a “supply chain risk” to national security, a first for an American company. In theory, it meant no organisation doing business with the US military would now be able to work with Anthropic. To top it off, Amodei’s nemesis and former employer, Sam Altman, had swooped in to announce a deal between his own company, OpenAI, and the Pentagon, offering its AI systems in classified operations in place of Anthropic’s.