Daniel Mercer
Reading breakup emails can feel intrusive, but when the parties involved include a billion-dollar AI lab and a major federal agency, curiosity is understandable. Recently released court documents shed light on the emails exchanged between Anthropic and the Department of Defense, detailing the fallout from their partnership earlier this year, and they confirm many of the public speculations surrounding the situation.
In these communications, Anthropic’s CEO Dario Amodei expressed his deep concern about the Department of Defense potentially utilizing his company’s AI models for applications that conflicted with his ethical standards. Specifically, he highlighted his unease regarding the integration of AI into autonomous weaponry and tools for domestic surveillance. The Pentagon’s stance, asserting that it could deploy the technology for “all lawful uses,” creates significant ambiguity in this context.
The origins of this rift can be traced back to January, when Emil Michael, the undersecretary of defense for research and engineering, contacted Amodei after a prolonged period of silence. In his message, Michael indicated he was “hoping that we are closer to engaging with your revised POV,” essentially expressing a desire for Anthropic to align with the Pentagon’s expectations. Amodei reiterated the necessity for safeguards in AI deployment, particularly against fully autonomous weapons and domestic surveillance applications.
Michael dismissed those concerns as “just not workable,” indicating there was “one more chance to align on core principles that would lead to legal language” before they would part ways. He emphasized that “there is no distinction in our world between weapons that are defensive or offensive,” highlighting the challenges Amodei would face in trying to draw a line regarding the potential applications of his technology.
Amodei pointed out that the Pentagon’s “all lawful uses” clause would not be acceptable for Anthropic, given that US laws permit domestic surveillance. Referring to some proposed language from the Pentagon, Amodei expressed to Michael that the department seemed to “completely remove our redlines.” The following day, Defense Secretary Pete Hegseth announced that Anthropic would be classified as a supply-chain risk, effectively terminating the negotiations.
Throughout the exchanges, Amodei maintained a strong stance on his principles, which reflects positively on him. Conversely, Michael appeared to compromise his integrity in attempts to secure the concessions the Department of Defense sought. His position was further complicated by his substantial stock holdings in xAI, a competitor of Anthropic, among other AI investments. Even if financial gain was not his primary motivation during negotiations with Amodei, his focus on maintaining avenues for military applications of AI raises ethical questions about his moral compass.
We have made the complete 346-page court document available for you to examine here.
Brandon Fletcher
