Anthropic’s Pentagon Problems
Plus, cleaning up cloud storage, Google’s TPU gambit, Meta goes on trial and a ‘Fitbit for farts’
Tensions between the Pentagon and Anthropic escalated in a big way this past week.
Officials at the Department of Defense were already considering canceling Anthropic’s contract due to the company’s reluctance to sign on to new terms. The Pentagon wanted the startup’s popular chatbot, Claude, to be available to the military for “all lawful uses,” we reported late last month.
Then things went nuclear, after we broke the news Anthropic had been used, via its partner Palantir, in the operation to capture Venezuelan President Nicolás Maduro.
Incensed that Anthropic had asked Palantir about how Claude was used in the raid, the Defense Department dialed up the heat. It threatened to label Anthropic—the only LLM currently cleared to work on classified material—a “supply chain risk,” the Journal reported. That designation is normally reserved for foreign actors like Huawei that would require the department to sever ties with the startup just a few months ahead of its expected IPO.
Anthropic, for its part, says its negotiations with the Pentagon continue to be productive, and the questioning about the Maduro raid was part of a routine discussion.
Even if the two sides manage to find a way to salvage their relationship, the question the spat raises about who should ultimately control AI—its creators, its users or the government—isn’t likely to be resolved anytime soon.