In the back-and-forth, Anthropic CEO Dario Amodei made clear that his primary concern was that the Defense Department would use his lab’s AI models for purposes with which he and the company were not comfortable — namely, integrating the technology into autonomous weapons systems and domestic surveillance devices. As mentioned earlier, the Pentagon’s position is that it can use the technology for “all lawful uses”, which leaves a considerable amount of leeway.
The emails show that the seeds of division were sown in January, when the Defense Department’s undersecretary for research and engineering, Emil Michael, reached out to Amodei after several weeks of radio silence. In it, Michael said he’s “hoping we’re closer to engaging with your revised POV” — basically, hoping Anthropic is willing to play along with the Pentagon’s demands. Amodei responded by reiterating his position that guardrails need to be in place for the use of AI, including prohibiting its use for fully autonomous weapons and domestic surveillance.
Michael said this was “absolutely not practical”, and warned that there was “another chance to agree on the basic principles that will give rise to the legal language” before deciding on separation. He further stated that “in our world there is no distinction between defensive or offensive weapons” (a description that seems remarkable given the ongoing debate over “defensive” weapons sales to Israel), so attempts by Amodei to distinguish how his company’s technology could be used would be futile.
Amodei explained that the “all lawful uses” standard would not apply to Anthropic because US law allows domestic surveillance. In response to some pointed proposed language from the Pentagon, Amodei told Michael that the department is “completely removing our redline.” The next day, Defense Secretary Pete Hegseth announced that his agency would designate Anthropic as a supply-chain risk, which almost ended that entire conversation.
If anything, Amodei looks great in the email, given that he stands by his convictions. Less influential was Michael, who made every effort to get his agency to make the desired concessions. But Michael already looked pretty compromised in this whole situation, given that he was sitting on a big pile of stock for Anthropic competitor XAI, in addition to other investments in AI firms. Even though enriching himself was only in the back of his mind during the back-and-forth with Amodei, at the front of his mind was figuring out how to keep the door open to using AI to kill people and conduct surveillance on civilians, so not a person with a strong moral compass by any means.
We’ve uploaded the entire 346-page court document here for you to browse.
<a href