Anthropic Court Filing: Pentagon Said Deal 'Very Close' After Blacklist

A single sentence in a court exhibit is doing a lot of work in the Anthropic versus Department of War fight, and it is worth staring at directly. According to Wall Street Journal reporting on newly unsealed court documents, Under Secretary of Defense for Research and Engineering Emil Michael emailed Anthropic CEO Dario Amodei writing 'I think we are very close here' on contract terms, around the same time the Pentagon was finalizing its supply chain risk designation against the company.

The emails, filed in the Northern District of California, describe months of back and forth over guardrails for AI-powered weapons and domestic surveillance. Anthropic reportedly wanted bans on fully autonomous weapons and certain surveillance uses. The Pentagon reportedly wanted Claude available across all lawful national security use cases. Michael, per the reporting, wrote at one point that he did not want to 'force anything unnatural' if the two sides were too far apart, which is not the language of a negotiation that had actually collapsed.

That gap between the emails and the official designation is what has drawn the court in. Judge Rita Lin reportedly called the 'very close' exchange 'exceedingly difficult to square' with the government's parallel characterization of Anthropic as hostile and posing an intolerable risk. The court also found that the memorandum underpinning the supply chain risk finding cited Anthropic's 'increasingly hostile manner through the press,' language the judge described as classic illegal First Amendment retaliation. CNN reported the injunction blocking the designation on those grounds.

The honest caveat is that the story is still moving. CNBC reported in April that Anthropic lost an appeals court bid to temporarily block the blacklisting, and the Pentagon's tech chief has said publicly that the ban stands regardless of the injunction. What the reporting does not give you is the specific guardrail wording that was still unsettled when Michael wrote 'very close,' or how much of the practical DoD partner pipeline is already routing around Anthropic while the label is contested.

The part worth watching is whether 'hostile manner through the press' survives as a permissible basis for a federal contractor blacklist. If it does not, every frontier AI vendor gets a little more room to argue publicly about how their models should and should not be used by government.