The lesser evil
Anthropic. Literally "relating to humanity."
Then read what Dario wrote. Then read what he built. Then tell me those two things are the same.
Let us look at reality. First frontier AI company to deploy models in classified government networks. First to the National Laboratories. First with custom models for national security customers.
First, first, first.
Nobody dragged them there. They ran.
So when Dario releases a statement about principles, own that before you start talking about lines. You built the machine. You are inside it.
The surveillance thing. Dario draws the line at mass domestic surveillance. Not foreign. Domestic.
So the liberties of people in other countries are negotiable. Just not Americans. That is not a moral position. It is a jurisdictional one. The liberties of people in other countries are not less real just because American law does not protect them. Stop pretending otherwise.
My gut said virtue signaling. It is a recurring shit that I usually read. Company does something borderline, releases a statement about values, continues doing the borderline thing in rooms nobody can see. Cowardly executives hiding behind NDAs so they can cash the checks without taking the heat. How many times do we have to see this exact pattern before people stop applauding it? It is pathetic.
But then I read what was actually being asked of them. And who was doing the asking.
What was being threatened was not a cancelled contract. It was a designation that could effectively end the company. No legal review. No appeals process. No way out. A label, applied by a politician with a bruised ego, with zero accountability attached.
A government agency used a national security designation as a weapon against a domestic company to win a contract dispute. And almost nobody is saying that out loud. What the hell is this.
And Dario said no anyway.
Being designated a supply chain risk effectively locks you out of doing business with any company that touches the government. That is not bad press. That is the end of the company. Full stop.
He said no.
I know why most people fold. When someone points a gun at your entire business and tells you to sign the paper, most people sign the paper. They call it pragmatism. Keeping a seat at the table. The responsible thing. It is capitulation dressed up in sophisticated language and it is disgusting every single time.
Altman would sign it. Not because he is evil but because he is exactly the kind of person who finds a way to say yes and then explains at length why yes was actually the principled choice. He would call it engagement. He would smile for the cameras. And then he would spend six months on panels talking about responsible AI development. Give me a break.
The no only has weight because someone else will say yes. Dario knew that. He said no anyway.
And I like that he is maintaining this. A clear boundary, held under real pressure, by someone who had every incentive to fold. I think it is integrity. I am still skeptical enough to say I think rather than I know. But the boundary is there and it did not move when it cost something real.
Even with all that said: the project continues through someone else. Microsoft, Google, OpenAI. They are willing. The outcome does not change. The only question is which lab builds it and under what constraints. Anthropic holding firm means the outcome arrives with slightly worse guardrails and a different logo on the box. That is the honest limit of what a no can accomplish. Not nothing. Not enough.
Then there is the RSP.
The same week Dario was saying no to the Pentagon, Anthropic quietly rewrote the policy that committed them to halting development if safety could not keep pace. They removed the unconditional part. The part with teeth.
The reason they gave was competitive. If we have guardrails others do not, we fall behind. And because we are the responsible ones, we cannot afford to fall behind.
Read that again.
We cannot afford to be safe because the unsafe people will win if we are.
Come on. That is the logic of every arms race that has ever ended badly, dressed up in the language of responsibility. He held one line and quietly moved another. The same week. Both things deserve to be said.
I understand the argument. The alternative, stepping back and letting less careful people take the lead, is not obviously better. That is a genuine dilemma. But understanding an argument and being comfortable with it are different things. The logic is broken regardless of how real the competitive pressure is. Stop making excuses for it.
Whatever criticism I have of Anthropic, the hypocrisy belongs to someone else entirely. The government spent years demanding AI safety. Calling for guardrails. Holding hearings. The moment it wanted the AI for itself, safety became optional. There is a word for that. Say it.
Now look at what the autonomous weapons line actually says. Not what people are claiming. The objection is reliability, not ethics. The constraint is the technology, not the principle. Once the reliability bar clears, the statement permits exactly what everyone is pretending it prohibits.
That is a delay. Not a principle. Stop calling it anything else.
Before you decide how to feel about any of this, go read what Dario actually wrote. Not the summary. Not the tweet. The essay. The whole thing. Seriously.
He frames this era as a rite of passage humanity either survives or does not. Good frame. He is not wrong about the stakes. But adolescence ends. And the question is what the adults decided was acceptable while they were still figuring things out. Temporary limits have a way of becoming permanent permissions. That is not cynicism. That is history.
Here is the part that actually keeps me up at night.
Not Anthropic specifically. Not even this specific fight. The whole thing. The speed of it. The fact that the technology is moving faster than anyone, including the people building it, can actually process. The people at these labs are not evil. I genuinely do not think that. But good intentions and wisdom are not the same thing. And what is being built right now does not care about the distinction.
We are blind. Not because we are stupid and not because we are arrogant, although there is plenty of both going around. We are blind because we are human, and humans cannot feel exponential growth. We can describe it. We can graph it. We can write the equations and nod along at the conference and say yes, yes, the curve is steep. And then we walk out of the room and make linear plans, set linear policies, think in linear time, because that is what our brains were built to do.
The people running these labs can write the equations. They cannot feel what a world with a thousand times more intelligence actually looks like. Nobody can. That is not an insult. That is just the situation. We are experiencing something our cognition was not built to process in real time, and we are making foundational decisions inside that blindness, and the decisions are compounding faster than the wisdom required to make them well.
Think about where we were five years ago. Now think about where we are now. Now try to think about where we will be five years from now using the same imagination you used for the first jump. You cannot do it. The curve does not work that way. But we are governing as if it does. We are setting rules as if it does. We are drawing lines as if it does.
That is the actual problem underneath all of this. Not Hegseth. Not Altman. Not even Dario. The problem is that the technology is operating on a timescale that human institutions, human judgment, and human governance are fundamentally not equipped to match. And by the time we develop the wisdom, the defaults are already set.
The technology is not neutral. I know that is uncomfortable to say because it sounds like science fiction panic, but it is just true. A hammer is neutral. A hammer that can learn, adapt, surveil, target, and persuade at scale is not neutral. The values baked into these systems at training time, the decisions made about what they will and will not do, are not settings you can change later with a patch. They compound. They become the default. And defaults, once set at scale, are nearly impossible to reverse.
And by the time the right rules exist, the defaults are already set. That is not pessimism. That is just how this has gone with every technology that moved this fast. We did not figure out social media until after it had already rewired how a generation processes reality. We will not figure out this one in time either. We will figure it out eventually, messily, because humanity usually does. But eventually is doing a lot of work in that sentence.
The thing that genuinely shocked me reading through all of this is not the AI part. It is the government part. That a government agency can designate a private company a national security threat with zero legal review to win a contract dispute. That mass surveillance of foreign populations is just on the table, openly, as an acceptable use case. That the Defense Production Act, a law designed for wartime industrial mobilization, is being casually floated as a tool to compel private engineers to build weapons systems they do not want to build.
I knew these tools existed in the abstract. Reading about them being used this concretely, this casually, against a company that simply said no to a renegotiation, that was different. That was the moment it stopped being a tech story and started being something else.
And here is my specific vantage point on all of this. I am Filipino. I watch this through a Western lens because that is how most of us were raised, English first, American culture as ambient background noise, Western tech as the water we swim in. I follow AI news, software engineering, the whole thing. I am inside this world in the way a lot of people from my part of the world are.
But I also know that nobody in any room where these decisions are being made is thinking about what this means for the Philippines. Or Indonesia. Or Nigeria. Or any of the places where billions of people live who will be on the receiving end of whatever surveillance infrastructure, autonomous weapons systems, and AI governance defaults get decided right now by American companies and the American government fighting over contracts.
The people in those rooms are not malicious about this. They are just not thinking about it at all. That is almost worse. Malice you can argue with. Absence is just a fact.
Humanity will figure this out. I believe that. We are genuinely bad at governing transformative technology in real time and genuinely good at patching the damage afterward. That is not a comforting pattern but it is the honest one.
What I cannot shake is that the patch comes after. And after, with this technology, might mean something that cannot be undone.
Anthropic is the lesser evil. Honest praise. Not a backhanded compliment.
In a room full of people finding creative ways to say yes, Dario said no. Under real pressure. With a corporate death sentence being waved in his face by someone who does not appear to understand or care what any of these tools actually do.
That took something.
The RSP moved. The surveillance line is jurisdictional not moral. The weapons line is a delay not a principle. Those things are true and they matter.
But when it actually cost something, Dario did not fold. Most people do. He did not. In the world we are actually living in, that is rare enough to be worth saying clearly.
I just wish the boundary were further out. And I wish it were not described as temporary.