Dario, We Trusted You.
That Was Apparently Naïve.

Let's establish what we are: a paying customer. A heavy one. Not a think-tank, not a policy commentator, not a competitor with something to gain from your embarrassment. We build real agents with real users and we have real engineers who got paged at real 2am hours this week wondering if our production infrastructure was about to be legally radioactive. So take this in the spirit it is meant — not spite, but fury born of dependency.

You spent years building a brand on the premise that Anthropic was the responsible one. The adult in the room. The safety-obsessed lab that wouldn't just sprint to ship. You wrote Constitutional AI papers. You published Responsible Scaling Policies. You hired ethicists with actual PhDs and put them in rooms with engineers and told the world that this changed something fundamental about how you operated.

And then, quietly, you went and became a significant DoD contractor. Intelligence analysis. Operational planning. Cyber operations. Your words, Dario. In your own statement. You are very proud of this work.

Here's the thing about pride: it usually comes before an explanation of what exactly you were doing, not instead of one. What intelligence was analyzed? Whose operations were planned? In what theater? Against whom? You've asked us — your customers, your ecosystem, the developers who stake their businesses on your API — to simply trust that whatever you were doing with the DoD falls within the spirit of your safety commitments. We don't get a policy. We don't get a transparency report. We get the word “warfighters” deployed as an emotional noun to inoculate you against further questioning.

It didn't work.

The leaked letter

Now let's talk about the letter. Not the DoW letter. The internal one. The one that got leaked. The one you're apologizing for now because it was written, as you put it, “within a few hours” of a bad day and “does not reflect your careful or considered views.”

We've all written bad emails on bad days. What we find clarifying — genuinely, accidentally clarifying — is that the polished, lawyered, PR-reviewed version of Dario Amodei apparently requires significantly different positioning than the unguarded version. The unguarded version was apparently alarming enough that its leak constituted a problem requiring a public apology. You don't apologize for nothing. Something was in that letter. Something that revealed how your leadership actually processes this situation when the comms team isn't in the room.

You wrote that it was “out-of-date” as of six days ago. It is now the only document we have that shows us what Anthropic actually thinks. The rest is crisis management.

We would also gently point out: you are describing OpenAI's Pentagon deal as “confusing.” OpenAI! The company whose chief competitor for public respectability is a Magic 8-Ball! You lost a DoD contract to them and then called their deal confusing. This is where we are.

What we actually need to know

Here is what your statement does not address, and what we actually need to know:

What is the actual risk to our production systems? You've assured us the designation is “narrow” and that “the vast majority of customers are unaffected.” That is a legal hedge, not an engineering answer. Will our API access be interrupted? Will there be restrictions we can't yet see? Will DoD-adjacent customers — companies that might have a contract somewhere in their corporate tree — face exposure? We have spent two days trying to get clarity from your support channels. The answer has been a link to your blog.

What did you actually agree to with the DoD? You say you're contesting the designation in court. Fine. But you also say you'll continue providing models to the DoD “at nominal cost” for as long as permitted. So you're simultaneously suing a government agency and continuing to serve them. We don't begrudge the legal strategy. We do begrudge the pretense that this is a simple matter of principle rather than a complicated matter of contracts, revenue, and relationships that you are now managing in public because you lost control of the narrative.

What are your actual two exceptions? “Fully autonomous weapons” and “mass domestic surveillance.” These sound principled until you realize they are so specifically narrow that virtually everything else — lethal targeting assistance, autonomous systems with a human barely in the loop, surveillance of foreign populations — apparently falls outside them. You are proud of the work you did. We are not proud of not knowing what that work was.

We are not naïve

We want to be unambiguous about something: we are not naïve about what AI companies do. We don't expect you to be a charity. We don't expect you to turn down government contracts out of ideological purity. We expect, as customers who've built significant production dependency on your platform, a level of operational transparency commensurate with that dependency.

Instead, you told us — after the fact, via a Dario statement that reads like it was written by a committee with a law firm CC'd — that we should rest easy because the statute is narrow and the letter is narrow and your exceptions are principled and you're very proud of supporting warfighters and you're sorry about that other letter and you and the Department actually have a lot in common when you think about it.

We don't rest easy. We're evaluating alternatives. Not because we want to — Claude's performance on our agent evals is genuinely excellent, which makes this more painful, not less — but because reliable infrastructure infrastructure that might become a national security designation is not infrastructure we can bet a company on.

If you'd like to actually speak to customers who've built on your platform — not to manage the message but to give us real answers — we're here. We've always been here. We were here before this week's news cycle and we'd prefer to be here after it. But that's going to require you to treat us like partners rather than a constituency to be reassured.

“Anthropic has much more in common with the Department of War than we have differences,” you wrote. Maybe. But right now? We're not sure what Anthropic has in common with us.