Anthropic – Pentagon emails expose the true conflict.

      For several months, the dispute between Anthropic and the Pentagon appeared to revolve around access to Claude. However, court documents released this week indicate a larger issue: who determines how the US military utilizes cutting-edge AI technology.

      The emails were disclosed on Tuesday as part of one of the lawsuits Anthropic has filed against the Department of Defense, with the Wall Street Journal being the first to report them. They reveal a tense back-and-forth between two individuals: Anthropic CEO Dario Amodei and Emil Michael, the undersecretary of defense for research and engineering.

      Amodei maintained a consistent stance, seeking restrictions on how the Pentagon could apply Anthropic's models. He excluded two specific uses: fully autonomous weapons and domestic surveillance, which were the same limits he previously outlined when addressing the military application of his models.

      In contrast, the Pentagon sought a broader scope, requesting that the models encompass “all lawful uses,” a term that provides substantial leeway.

      Tensions began to escalate in January when Michael reached out to Amodei after weeks of silence, expressing a hope that they were nearing agreement on Amodei's revised perspective. Essentially, he wanted Anthropic to align more closely with the Pentagon's viewpoint, but Amodei reiterated his limitations.

      Michael was straightforward, stating that the proposed guardrails were “just not workable.” He offered Anthropic “one more chance to align on core principles” before they parted ways, also dismissing the boundaries set by Amodei, asserting, “There is no distinction in our world between weapons that are defensive or offensive.” Gizmodo published the emails from the court records.

      Amodei challenged the “all lawful uses” criterion, pointing out that US law does allow for domestic surveillance. He conveyed to Michael that the Pentagon's proposed language extended too far, effectively “completely removing our redlines.”

      The following day, Defense Secretary Pete Hegseth labeled Anthropic as a supply-chain risk, a designation typically reserved for companies associated with foreign adversaries, escalating the months-long deadlock into a legal battle.

      Michael's role has come under scrutiny due to financial disclosures revealing his stock ownership in xAI, a competitor of Anthropic, as well as other AI investments, as reported by The Guardian and ProPublica. Anthropic contends that the blacklisting was an act of retaliation rather than a security measure, a claim that a federal judge supported in late March by granting a preliminary injunction, characterizing the action as “classic illegal First Amendment retaliation.” An appeals court overturned this decision in April, and the case is ongoing. This dispute exemplifies the growing connections between Silicon Valley and the Pentagon.

      The implications extend beyond Washington. Anthropic is testing whether an AI company can impose ethical boundaries on a governmental customer while retaining its contracts. This issue is central to Europe’s ongoing discussions regarding military and surveillance AI, as embodied in the EU AI Act, which is grappling with similar challenges. The conversation also fuels the sovereignty debate as European buyers consider the level of control any US laboratory maintains once its models are deployed in national security applications.

      Currently, Anthropic has managed to stabilize its position, having made progress in resolving its recent conflict with the administration and nearing an agreement to regain access to a restricted model. The emails illustrate how precariously close the relationship came to breaking and the reasons behind it.