According to TechCrunch, artificial intelligence company Anthropic officially submitted a sworn statement to the California federal court to refute the Pentagon’s accusation that it poses a “national security risk.” This lawsuit, triggered by the U.S. government's unilateral announcement of severing cooperation, is revealing more details. The latest court documents show that the two sides were actually very close to reaching a consensus before a complete break.
According to documents filed by Anthropic policy director Sarah Heck, concerns that the Pentagon asserted in court that Anthropic was seeking approval authority for military operations and that the technology could be deactivated mid-operation were never raised in negotiations in the months before the dispute. What’s even more dramatic is that the day after the Department of Defense officially listed the company as a supply chain risk (March 4), Deputy Secretary of Defense Emil Michael sent an email to Anthropic’s CEO, making it clear that the two sides were “very close” to an agreement on the two core points of disagreement between autonomous weapons and mass surveillance of the American people. This is in stark contrast to the tough stance the U.S. government subsequently conveyed to the public.
In response to technical security concerns, Anthropic public sector director Thiago Ramasamy made a technical refutation in a statement. He pointed out that once the large AI model Claude is deployed in a government system operated by a third-party contractor, Anthropic does not have any access rights, there is no remote kill switch or backdoor, and it is technically impossible to intervene in military operations. In addition, in response to allegations of risks caused by hiring foreign employees, the document emphasizes that Anthropic’s employees involved in the construction of confidential environment models have passed security clearance reviews by the U.S. government.
Currently, Anthropic insists in the lawsuit that this supply chain risk identification, the first in U.S. history for a local company, is essentially the government’s First Amendment retaliation against the company for publicly expressing its views on AI safety. The U.S. government denied this in a 40-page document earlier this week, arguing that the exclusion of Anthropic was based on a purely national security decision and was not punishment for his speech. A hearing on the case will be held on March 24 in San Francisco.