Why AI won’t go well unless sensible people like you speak up and act.
A response to recent events.
The US Secretary of War, Pete Hegseth, just announced that the DoW will label Anthropic a supply chain risk for refusing to remove contract terms that prohibit the DoW from using their AI for domestic mass surveillance and lethal autonomous weapons.
Yup, you read that right.
Quick tldr on Anthropic vs. DoW
Anthropic and the DoW had a contract that allowed for all legal use of the AI models that included prohibitions on the above two items.
The DoW wants to remove these two prohibitions from the contract, but Anthropic refused and stood their ground.
Instead of going the ‘normal’ route of cancelling the contract and working with other labs like XAI, Google, or OpenAI, Hegseth announced that he is moving to designate Anthropic as a supply chain risk. This measure is usually reserved for companies with foreign adversary ties like Huawei and Kaspersky Lab. Legally, this means firms cannot use Anthropic products to fulfill DoW contracts. However, Trump and Hegseth suggest that no contractor can do business with Anthropic at all.
Just hours after Hegseth’s announcements on Friday, Sam Altman announced that OpenAI “Reached an agreement with the Department of War to deploy our models in their classified network.” despite allegedly saying1 they “will draw the same red lines” in an internal memo.
Note: XAI has already complied with the DoW days prior.
Wow what the f*ck just happened
The situation is ongoing; I do not claim to know the full picture. I’m also not trying to convince you of anything on this matter. You should consult trustworthy sources to make up your own mind (see Anti-DoW view, neutral view, Pro-DoW view).
Before I give any takes, I need to say that I’m biased. I do research and strategy work to ensure that AI is developed and used for humanity’s benefit, and as far as I can tell, Anthropic is the only frontier lab that is actually trying hard to accomplish this. They’re far from perfect2, but they’re the best frontier AI lab by far. This much seems obvious to me from personally speaking to their employees and from the quality of their work on technical AI safety.
First, I think we should all agree that the USG has the right, given by its people, to decide how AI is developed and deployed in the USA via legal and democratic processes (ie. through Congress). No private entity or person should decide this, certainly not Anthropic.
The DoW wants a contract that allows for “All legal use” of AI models, claiming that they have no intention of conducting mass surveillance or building lethal autonomous weapons. Okay, so what’s the problem? From what I could find, both domestic mass surveillance and lethal autonomous weapons sit in gray areas in US law, there is patch work, but no concrete laws or definitions. I’m not a legal expert, please tell me if I’m wrong.
Domestic mass surveillance
The Fourth Amendment protects against unreasonable searches and seizures, though what counts as a “search” in the digital age is contested.
Under current law, the government can purchase detailed records of Americans’ movements, web browsing, and associations from public sources without obtaining a warrant.
Lethal autonomous weapons
DoD Directive 3000.09 requires “appropriate levels of human judgment” for autonomous weapons. But this is a policy directive, not a statute. It can be revised or waived by the Secretary of War - Hegseth himself.
The administration can change DoD directives at will.
Anthropic/Dario probably judged that they weren’t willing to risk Claude being used for such morally dubious things and stood firm on their principle. The people sure seem to love this. From personal anecdotes, an important decision factor was that Anthropic would’ve faced mass talent exodus if they complied with the DoW. A good portion of the lab’s top talent are effective altruists (aka EAs), who, if nothing else, are known for strong ideological stances. This is the same movement that many on the tech right seem to be trying to smear, this is what Hegseth had to say about them, make of it what you will.
Potential interpretations of the DoW’s seemingly contradictory actions I’ve seen:
Taking what the DoW says at face value - making sure “America’s warfighters will never be held hostage by the ideological whims of Big Tech.” Seems plausible, very reasonable, this should be how things are.
Revenge and ego - apparently, Anthropic inquiring Palantir about Claude’s use in Maduro’s capture triggered all of this. They want to send the message “build the best model, hand over the keys, and never ask what they did with it.”
Aiding an effort to take down Anthropic for the business interests of the administration’s tech right supporters under an ideological guise. VC firm a16z (no stake in Anthropic, major stake in OpenAI & SpaceX), and Greg Brockman (OpenAI President) launched a pro-AI superPAC not long ago. This is what they have to say about Anthropic. This is a whole rabbit hole in itself, here is a deep research report3 using Google’s AI Gemini.
They actually want to conduct mass surveillance and build lethal autonomous weapons - I sure hope not!
You should choose what you want to believe based on available evidence, and definitely check out what other people have said.
Silence is complicity
This is the symptom of a wider, systemic problem: none of our institutions are prepared for the speed and significance of AI development. Risks of misalignment and “Artificial Super Intelligence” aside, at the current rate, we almost certainly do not have enough time to react and adapt to how AI will change international relations, domestic production, and national security. Yet, the average person outside of tech doesn’t really know what AI is, and the average AI/ML researcher doesn’t seem to care about events like the Anthropic & DoW feud. We desperately need leaders and citizens across sectors to educate themselves on what AI is, how it might impact crucial functions they perform, and act on it. The water is muddy because most people aren’t trying to make it cleaner. AI progress seems to be exponential, yet we operate at human speed, and democracy is even slower. We need to change that.
It frustrates me greatly when I tell my very smart and capable friends about such problems, encourage them to work on it, and the average reaction I get is basically “seems good to do, I hope you succeed, I’m gonna move on with my life”. This isn’t even an appeal to moral duty - if AI systems of the immediate future are not developed in and governed by a stable democracy, it probably wouldn’t be used in your best interests, and you will likely not enjoy such a world. To be clear, I am uncertain. I don’t know what the future holds, and I often question whether I’m the fool for feeling so strongly about these issues. But each time I look harder, evidence seems to vindicate me. If I know you in real life, we may have had this conversation, and I’m sincerely sorry for any discomfort I may have caused, but please hear this.
Yes, we are just students trying to find a job and make a living, there’s absolutely nothing wrong with that. Yes, the CS job market is cooked. It can make us feel powerless, like the only thing we can do is fend for ourselves and nothing more. But don’t let this disempower you, don’t accept it. Learned powerlessness is real, the most valuable asset in a world with AI is human agency, so don’t lose it. Look for real problems, and try real hard at solving them. You will be able to find a job and make a great living if you meet a genuine need. This is how capitalism is supposed to work.
“If I don’t do it, someone else will” is a dangerous way to think. Moral abdication is what leads to the fall of empires, and the root cause of all arms races. We live in a self fulfilling prophecy, a world where all frontier AI labs face the same mass exodus pressure as Anthropic is achievable if and only if you believe in it. It’s not the incentives, it’s you.
The US seems like our best shot at developing frontier AI in a stable democracy. If you’re a US citizen, you have the rare privilege of living in a democracy. We have a shared responsibility to educate ourselves, see through the circus, consider opposing perspectives, and make up our minds. We are witnessing for the first time an AI lab, an entity not elected by the people, throw its weight around against a nation state - this should concern you. We desperately need governmental capacity to hear what the people want and act on such issues. People much less sensible than yourself without your best interests in mind are trying very hard to influence the course of AI development. You should get off your ass and do something.
Here’s what you can do
Educate yourself4, refer to trusted sources, present reasoned arguments in good faith. Make yourselves heard, call you representatives. This is the very foundation of democracy.
If you’re a consumer, vote with your actions.
If you’re a developer/researcher, vote with your feet.
If you work at Google/OpenAI, sign the We Will Not Be Divided Open Letter.
If you want to pursue an impactful career that tackles such problems from AI, talk to me, I want to help.
This was a long, rambly, emotionally charged blog - not what I usually write. I pulled an all-nighter and I do not enjoy writing like this - had to adopt too much of a soldier mindset. I sincerely hope the point gets across, thank you for hearing me out.
I don’t trust Sam’s words as a policy. See this Midas Project article.
Despite what this post may read like, I’m not trying to advocate for or defend Anthropic. Objectively, they are a for-profit AI lab that has accelerated the race dynamic. They do sneaky things like dropping their safety pledge from the Responsible Scaling Policy and using the positive PR of the DoW situation as cover. I’m unsure whether a rational actor that truly cares for AI to go well would walk back their safety policy and I and remain skeptical. On one hand it’s a race to the bottom, on the other, to exert influence over the world and do safety research, you need massive amounts of resources and frontier AI models. I probably disagree with Anthropic’s decision. I think they should stick to the if-then commitments and be very loud about it when they cross their own red lines and stop releasing models. If this situation showed anything, it’s that people care.
This is the prompt I used for Gemini: “One possible take on the DoW's recent confusing actions against anthropic is that there is a scheme by the tech right to take down anthropic for business reasons using ideological mask as disguise. Conduct a thorough investigation and only present the facts. Find relevant connections/histories of Pete Hegseth & the Trump administration and any of his key informants/decision makers, and whether they are connected in any way to Anthropic's business rivals.”
To learn more about AI, it’s rate of progress, and risks we may be concerned about. Check out BlueDot Impact’s courses and chapters 1 and 2 of the AI Safety, Ethics, and Society Book, they’re both free.
This is really good, and matches exactly what I’ve seen trying to push AI safety as a concern for right now versus “we’ll deal with it when we get there”. The only pushback I had was in the footnotes - I am not necessarily convinced that Anthropic has materially accelerated the race in a way that US-China dynamics and Google v OpenAI would not/were not doing themselves. In light of that, I’d much rather have a competitive model by the company that seems to most consider risk, and I worry that that is a orthogonal goal to being able to take as unilateral a safety stance as they initially had.
With that said, I share many of the same biases regarding anthropic you mentioned, so take that for what it’s worth, and the rest was a great view into a problem that needs more reasoned analysis than it’s currently getting.