Trump administration doubles down on Anthropic blacklisting in court arguments

"The Pentagon claims it's unworkable for the military or its vendors to rely on Anthropic because the company might pull the plug at any time due to its "ideological" views around AI safety. Unlike other model-makers, Anthropic refused to agree to the Pentagon's "all lawful use" standard for AI deployment. Anthropic argues it has no way to control its AI models once they're deployed in classified settings, and has stuck to its red lines around the use of its tools for mass domestic surveillance or the development of weapons that fire without human involvement."

""For the life of me, I do not see any evidence of maliciousness despite the best efforts of [Pentagon Under Secretary Emil Michael], who in his memo refers to you as having mal-intent, a bad motive, cannot be trusted," DC federal appeals court Judge Karen Henderson said. "To me this is just a spectacular overreach by the department.""

"In an exchange with Anthropic's lawyer, Judge Gregory Katsas pinpointed the difficulty of the company's usage policies for evolving AI models. "It doesn't really matter whether we focus on what might happen with the one they're currently using or what might happen with the one that everyone knows they will need three months from now, because AI three months from now will be totally different from the AI of today." Katsas and Judge Neomi Rao raised the Pentagon's concern that Claude is opaque (a feature of AI models that is not unique to Anthropic)."

""Forget about all the crazy rhetoric about... 'crazy company' ... and forget about the concern about the kill switch in real time. They still have this concern, right? The model is unpredictable," Katsas said. The other side: The government's lawyer, Sharon Swingle, suggested the main issue"

The Pentagon claims relying on Anthropic is unworkable because the company could stop providing models at any time due to ideological views on AI safety. Anthropic declined to accept an “all lawful use” standard for AI deployment. Anthropic argues it cannot control its models once deployed in classified environments, and it maintains restrictions against mass domestic surveillance and against weapons that fire without human involvement. During court arguments, judges questioned the Pentagon’s evidence and characterized the department’s position as an overreach. The court also focused on how evolving AI models make future behavior hard to predict, and on model opacity as a challenge for oversight. The government emphasized unpredictability as a core concern.

Read at Axios