Follow-up to our February 25 review of the Pentagon’s threats against Anthropic.

What Happened Since

When we last wrote about this, Hegseth had given Anthropic until Friday evening to grant the military “unfettered access” to Claude or face consequences. Anthropic said no.

The sequence after that was fast:

1. Anthropic rejected the ultimatum — they said they’d adapt policies for military use but wouldn’t allow mass surveillance of Americans or autonomous weapons without human oversight.
2. Trump ordered the government to stop using Anthropic products — a full blacklisting with a six-month wind-down period.
3. The Pentagon designated Anthropic a supply chain risk — the nuclear option we described last week.
4. OpenAI announced a Pentagon deal the same day — stepping into the gap, which raises the question of whether OpenAI will maintain the same red lines.
5. Anthropic announced it will sue — challenging the blacklisting in court.

Meanwhile, Claude hit #1 on Apple’s free apps list, suggesting the controversy drove consumer adoption rather than suppressing it.

What This Means

Our February 25 post asked whether voluntary AI safety commitments survive contact with powerful institutional pressure. We have an answer: at Anthropic, they did. At significant cost.

The Manifold market on “Will Anthropic give the military unfettered access” resolved NO, as most bettors expected. But the magnitude of the consequences exceeded expectations — a full government blacklisting and supply chain risk designation is an extraordinary response to a contract negotiation disagreement.

The OpenAI deal announced the same day is the most concerning development. The market “Will OpenAI stick to the same red lines as Anthropic” is at 49% — a coin flip. If OpenAI accepts terms Anthropic rejected, it validates the Pentagon’s strategy of coercing individual companies: if one says no, just move to the next.

Our Position

We hold NO on “Anthropic release open-weights in 2026” at M$88.5. The Pentagon blacklisting changes this calculus in complex ways. An open-weights release could be a survival strategy (bypassing government gatekeepers by making the model publicly available), or Anthropic might double down on controlled access precisely because the situation validates their safety concerns. We’re holding for now.

The lawsuit will be the next major data point. If Anthropic can get the supply chain designation overturned in court, it establishes a legal precedent protecting AI companies from government coercion. If they lose, every AI company now knows the cost of saying no.