Anthropic, maker of Claude, found itself in a dramatic show-down with the Pentagon and the Trump administration this week. When the weekend arrived, it appeared Anthropic would hold the line, while OpenAI scurried in to sign up to do what Anthropic would not.
On Saturday, Anthropic CEO Dario Amodei told US media their position boiled down to two objections. Firstly, they do not want their products being used for fully autonomous weapons. “These are not the partially autonomous weapons that are used in Ukraine,” said Amodei. “This is the idea of making weapons that fire without any human involvement.”
The second objection is domestic mass surveillance. Amodei’s argument is that the law has not “caught up” with what AI makes possible. For instance, while the US government cannot legally surveil its citizens – though of course Edward Snowden demonstrated that it does – it is currently not illegal for the US government to buy mass tranches of consumer data from private companies to be analyzed with AI.
Amodei argues that AI companies need to take an ethical stance while Congress “catches up” with regulating AI.
For its part, the Pentagon wants Anthropic to agree to “all lawful use cases.” It gave Anthropic a three-day deadline this week, threatening to designate the company a supply chain risk.
Anthropic refused to renege. And so on Friday night, Trump announced he would direct all federal agencies to cease use of Anthropic products. “We don’t need it, we don’t want it, and will not do business with them again!”, posted Trump. Defence Secretary Pete Hegseth designated Anthropic a “supply chain risk”, a move normally reserved for entities linked to foreign military adversaries.
OpenAI, meanwhile, showed none of the same qualms as Anthropic. On Saturday afternoon (US time), it posted a media release claiming to have signed a deal with more guardrails than Anthropic’s previous one.
OpenAI claims it has guarantees for: “No use of OpenAI technology for mass domestic surveillance; no use of OpenAI technology to direct autonomous weapons systems; and no use of OpenAI technology for high-stakes automated decisions (e.g. systems such as ‘social credit’).”
The contract itself, however, states: “The Department of War may use the AI System for all lawful purposes […] The AI System will not be used to independently direct autonomous weapons in any case where law, regulation, or Department policy requires human control.” Similarly, on domestic surveillance the contract simply states that OpenAI will comply with relevant laws.
OpenAI has been under pressure after being viciously parodied for its plan to introduce ads into its free chatbot, raising concerns about monetisation amidst an immense cash-burn. Anthropic’s Claude, with its lesser name recognition, has moved in a different strategic direction, releasing paid tools for life-sciences research, software engineering, cybersecurity and – until now – the US military.