“The World Is a Complicated, Messy, and Sometimes Dangerous Place”
Yesterday’s piece ended with a question: what else needs to be written down, and who else gets to make that demand? Here is what happened next.
After the deadline passed, Hegseth formally designated Anthropic a supply chain risk. Hours later, OpenAI took the contract.
OpenAI wanted the contract. They got it. They also wanted to look like they stood on principle. Sam Altman’s post on X is doing that work. It is caving with better optics. The X post is the needle-threading — enough public principle to satisfy the audience that was watching, enough flexibility in the contract language to satisfy the government that was paying.
What OpenAI did
In a post on X last night, Altman named the same two limits Anthropic was fired for insisting on. No domestic mass surveillance. No autonomous weapons. He said the Pentagon agrees with those principles, that they are “reflected in law and policy,” and that he is asking the Pentagon to offer the same terms to every AI company.
No contract has been signed. And “reflected in law and policy” is the Pentagon’s own argument from all week: mass surveillance is already illegal, so you don’t need it written down. The contract could read exactly as the government always wanted, with no written limits on either use, and the law interpreted however enforcement chooses. What Altman put in an X post and what ends up in executed contract language are two different things. One is a public statement. The other is what can be held to.
Altman told OpenAI employees at an all-hands yesterday afternoon that a potential agreement was emerging. An OpenAI official at that meeting said the Anthropic relationship broke down partly because Amodei had offended Department of War leadership by publishing blog posts “the department got upset about.” That framing positions OpenAI as the reasonable partner and Anthropic as the difficult one. It also has nothing to do with the substance of what either company was asking for.
What the timing means
Another important update is that this is no longer a hypothetical. Trump is calling for regime change. The hypothetical nuclear attack scenario that reportedly escalated the Anthropic negotiations is no longer a thought experiment. The U.S. is conducting active military operations. OpenAI’s models are now headed into the classified network where those operations are planned and executed.
The Pentagon spent a week insisting it had no interest in mass surveillance or autonomous weapons. It said those things were already illegal. It said trust us. And then it designated the one company that asked for that trust in writing a threat to national security, on the same night the United States and Israel began bombing Iran.
Altman’s post ends: “The world is a complicated, messy, and sometimes dangerous place.” He is not wrong. It is also the night the U.S. started bombing another country. The written limits he named in his post are, so far, a post.
What Anthropic did
Anthropic said it will challenge the supply chain risk designation in court. “No amount of intimidation or punishment from the Department of War will change our position on mass domestic surveillance or fully autonomous weapons,” the company said. Amodei told CBS News: “We are patriotic Americans. Everything we have done has been for the sake of this country, for the sake of supporting U.S. national security.”
Anthropic also pushed back on the scope of Hegseth’s designation, arguing he does not have the statutory authority to bar all Pentagon contractors from doing business with the company. The designation, they said, applies only to the use of Claude in Department of War contracts, not to how contractors use Claude to serve other customers.
That is a harder line than anything in Altman’s post. It is also the line of a company that no longer has the contract. Whether it holds in court is a different question than whether it held yesterday. It held yesterday. The cost was real.
What this means
The divide-and-replace strategy worked. Anthropic was replaced. The government got into the classified network it wanted, with a company willing to sign on terms it can interpret as it chooses, on a night it is using military force in the Middle East.
Yesterday’s piece argued that written specificity is the only thing that can be held to. That argument is not weaker this morning. It is more urgent. OpenAI is now inside, with limits named in public and not yet written into an executed contract, while the U.S. conducts active military operations in which these systems will be used.
The world is complicated, messy, and sometimes dangerous. It is also the world in which these systems are operating right now, not in a hypothetical, not in a negotiation, but in actual military operations. What is written down matters. What is not written down matters more.
Here is our post from yesterday:
Anthralytic is a strategy and evaluation studio that helps mission-driven teams clarify and amplify their impact.
