Emails reveal Anthropic's standoff with Pentagon over autonomous weapons and surveillance AI

Read the Tense Emails Between the Pentagon (Former Uber Exec) and Anthropic (Dario Amodei)

It feels voyeuristic to read someone's breakup texts, but usually the parties involved in the split aren't an AI lab with a billion-dollar valuation and the biggest agency in the federal government, so you can be forgiven for being interested in taking a peek. On Tuesday, court documents were released that include the emails that led to Anthropic and the Department of Defense falling out earlier this year, and they read just about as you'd expect based on the public accounting of the situation. In the back-and-forth, Anthropic CEO Dario Amodei made clear that his primary concern was that the Department of Defense would use his lab's AI models for purposes that he and the company weren't comfortable with -- namely, integrating the technology into autonomous weapons systems and domestic surveillance tools. As has been noted previously, the Pentagon's position is that it could utilize the technology for "all lawful uses," which creates a considerable amount of wiggle room. The emails do reveal that the seeds for the split were planted in January, when the Department of Defense's undersecretary of defense for research and engineering, Emil Michael, reached out to Amodei after radio silence for several weeks. In it, Michael said he was "hoping that we are closer to engaging with your revised POV" -- basically, hoping that Anthropic is ready to play ball with the Pentagon's demands. Amodei responded by reiterating his position that there need to be guardrails in place for the use of AI, including barring its use for fully autonomous weapons and domestic surveillance. Michael said that was "just not workable," and warned there was "one more chance to align on core principles that would lead to legal language" before deciding to part ways. He followed up by noting that "there is no distinction in our world between weapons that are defensive or offensive" (a detail that seems noteworthy given the ongoing debate over the sale of "defensive" weapons to Israel), so attempts by Amodei to make a distinction as to how his company's technology could be used would be futile. Amodei pointed out that the "all lawful uses" standard wouldn't fly for Anthropic because US law does allow for domestic surveillance. In response to some apparent proposed language from the Pentagon, Amodei noted to Michael that the department seemed to "completely remove our redlines." The next day, Defense Secretary Pete Hegseth announced that his agency would be designating Anthropic as a supply-chain risk, which pretty much put an end to that whole negotiation. If anything, Amodei comes off looking pretty good in the emails, given that he held to his convictions. Less impressive was Michael, who did everything he could to extract the concessions his agency was looking for. But Michael already looked pretty compromised in this whole situation, given that he was sitting on a fat stack of stock for Anthropic competitor xAI, among other investments in AI firms. Even if enriching himself was only in the back of his mind during his back-and-forth with Amodei, at the front of his mind was figuring out how to keep the door open for using AI to kill people and surveil citizens, so not exactly a guy with a strong moral compass either way. We've uploaded the full 346-page court document for you to browse here.

Anthropic - Pentagon emails reveal the real fight

Court documents released this week lay bare the back-and-forth between Anthropic's Dario Amodei and Pentagon undersecretary Emil Michael. The real fight was not over access to Claude, but over whether an AI lab can bar the US military from using its models for autonomous weapons and domestic surveillance. For months, the fight between Anthropic and the Pentagon looked like a row over access to Claude. Court documents released this week suggest it was about something bigger: who decides how the US military uses frontier AI. The emails came out on Tuesday, in one of the lawsuits Anthropic has filed against the Department of Defense. The Wall Street Journal first reported them. They trace a tense exchange between two men. One is Anthropic chief executive Dario Amodei. The other is Emil Michael, the undersecretary of defence for research and engineering. The redline Amodei's position stayed consistent. He wanted guardrails on how the Pentagon could use Anthropic's models. Two uses were off the table: fully autonomous weapons and domestic surveillance. Those are the same limits he set out when he questioned the military use of his models. The Pentagon wanted something broader. It asked for the models to cover "all lawful uses," a phrase that leaves plenty of room. The talks began to sour in January. Michael emailed Amodei after weeks of silence. He said he was "hoping that we are closer to engaging with your revised POV." In plain terms, he wanted Anthropic to shift towards the Pentagon's position. Amodei repeated his redlines instead. 'Just not workable' Michael was blunt. The guardrails were "just not workable," he wrote. He offered Anthropic "one more chance to align on core principles" before the two sides split. He also rejected the line Amodei drew. "There is no distinction in our world between weapons that are defensive or offensive," he wrote. Gizmodo published the emails from the court record. Amodei pushed back on the "all lawful uses" standard. The problem, he noted, is that US law does permit domestic surveillance. He told Michael the Pentagon's draft language went too far. It seemed to "completely remove our redlines." The next day, Defence Secretary Pete Hegseth designated Anthropic a supply-chain risk. That label usually applies to firms tied to foreign adversaries. It turned a months-long standoff into a court fight. A conflicted negotiator Michael's role has drawn scrutiny. Financial disclosures show he held stock in xAI, an Anthropic rival, alongside other AI investments. The Guardian and ProPublica reported the holdings. Anthropic argues the blacklisting was retaliation rather than security policy. A federal judge agreed in late March. She granted a preliminary injunction. She called the move "classic illegal First Amendment retaliation." An appeals court reversed that in April, and the case continues. The dispute is one strand in Silicon Valley's deepening ties to the Pentagon. Why Europe should watch The stakes reach past Washington. Anthropic is testing whether an AI company can set ethical limits on a government customer and keep its contracts. That question sits at the centre of Europe's own arguments over military and surveillance AI. The EU AI Act is wrestling with the same lines. It also feeds the sovereignty debate, as European buyers weigh how much control any US lab keeps once its models enter national-security work. For now, Anthropic has steadied itself. It overcame its latest clash with the administration this week, nearing a deal to restore access to a restricted model. The emails show how close the relationship came to breaking, and why.