OpenAI has the same redlines as Anthopic based on Altman's statements [2]. However somehow Anthropic gets banished for upholding their redlines and OpenAI ends up with the cash?
[1]: https://xcancel.com/OpenAI/status/2027846013650932195#m
[2]: https://www.npr.org/2026/02/27/nx-s1-5729118/trump-anthropic...
When Anthropic says they have red lines, they mean "We refuse to let you use our models for these ends, even if it means losing nearly a billion dollars in business."
When OpenAI says they have red lines, they mean "We are going to let the DoD do whatever the hell they want, but we will shake our fist at them while they do it."
That's why they got the contract. The DoD was clear about what they wanted, and OpenAI wasn't going to get anywhere without agreeing to that. They're about as transparent as Mac from It's Always Sunny in Philadelphia when he's telling everyone he's playing both sides.
Very gracious of OpenAI to say Anthropic should not be designated a supply chain risk after sniping their $200 million contract by being willing to contractually let the government do whatever they like without restrictions.
https://openai.com/index/our-agreement-with-the-department-o...
The current administration is so incompetent that I find this perfectly believable.
I imagine the government signed with OpenAI in order to spite Anthropic. The terms wouldn't actually matter that much if the purpose was petty revenge.
I don't know if that's actually what happened here, I just find it plausible.
Anyway, it is also amusing to hear tech people defend their right to earn some of the fattest salaries on this planet using the smol bean technique after a decade of "why wouldn't the West Virginian coal miner just learn to code." It was always about maintaining the lifestyle of yearly Japan vacations and MacBook upgrades and never about subsistence.
Except they are not "more stringent".
Sam Altman is being brazen to say that.
In their own agreement as Altman relays:
> The AI System will not be used to independently direct autonomous weapons in any case where law, regulation, or Department policy requires human control
> any use of AI in autonomous and semi-autonomous systems must undergo rigorous verification, validation, and testing
> For intelligence activities, any handling of private information will comply with the Fourth Amendment, the National Security Act of 1947 and the Foreign Intelligence and Surveillance Act of 1978, Executive Order 12333, and applicable DoD directives
> The system shall also not be used for domestic law-enforcement activities except as permitted by the Posse Comitatus Act and other applicable law.
I don't think their take is completely unreasonable, but it doesn't come close to Anthropic's stance. They are not putting their neck out to hold back any abuse - despite many of their employees requesting a joint stand with Anthropic.
Their wording gives the DoD carte blanch to do anything it wants, as long as they adopt a rationale that they are obeying the law. That is already the status quo. And we know how that goes.
In other words, no OpenAI restriction at all.
That is not at all comparable to a requirement the DoD agree not to do certain things (with Anthropic's AI), regardless of legal "interpretation" fig leaves. Which makes Anthropic's position much "more stringent". And a rare and significant pushback against governmental AI abuse.
(Altman has a reputation for being a Slippery Sam. We can each decide for ourselves if there is evidence of that here.)
As Paul Graham said, "Sam gets what he wants" and "He’s good at convincing people of things. He’s good at getting people to do what he wants." and "So if the only way Sam could succeed in life was by [something] succeeding, then [that thing] would succeed"
--Paul Graham, 2008
That's not quite right.
First off, I don't expect that "you used my service to commit a crime" is in and of itself enough to break a contract, so having your contract state that you're not allowed to use my service to commit a crime does give me tools to cut you off.
Second, I don't want the contract to say "if you're convicted of committing a crime using my service", I want it to say "if you do these specific things". This is for two reasons. First, because I don't want to depend on criminal prosecutors to act before I have standing. Second, because I want to only have to meet the balance of probabilities ("preponderance of evidence" if you're American) standard of evidence in civil court, rather than needing a conviction secured under "beyond a reasonable doubt" standard. IANAL, but I expect that having this "you can't do these illegal things except when they aren't illegal" language in the contract does put me in that position.
They literally asked the DoD to continue as is.
Their is no safety enforcement standing created because their is no safety enforcement intended.
It is transparently written, as a completely reactive response to Anthropic’s stand, in an attempt to create a perception that they care. And reduce perceived contrast with Anthropic.
If they had any interest in safety or ethics, Anthropic’s stand just made that far easier than they could have imagined. Just join Anthropic and together set a new bar of expectations for the industry and public as a whole.
They could collaborate with Anthropic on a common expectation, if they have a different take on safety.
The upside safety culture impact of such collaboration by two competitive leaders in the industry would be felt globally. Going far beyond any current contracts.
But, no. Nothing.
Except the legalese and an attempt to misleadingly pass it off as “more stringent”. These are not the actions of anyone who cares at all about the obvious potential for governmental abuse, or creating any civil legal leverage for safe use.
It's even worse than that, because this administration has made it clear they will push as hard as possible to have the law mean whatever they says it means. The quoted agreement literally says "...in any case where law, regulation, or Department policy requires human control" - "Department policy" is obviously whatever Trump says it is ("unitary executive theory" and all that), and there are numerous cases where they have taken existing law and are stretching it to mean whatever they want. And when it comes to AI, any after-the-fact legal challenges are pretty moot when someone has already been killed or, you know, the planet gets destroyed because the AI system decide to go WarGames on us.
The Trump administration acts cartoonish and fickle. They can easily punish one group, and then agree to work with another group on the same terms, to save face, while continuing to punish the first group. It doesn't have to make consistent sense. This is exactly how they have done with tariffs for example.
Secondly, the terms are technically different because "all lawful uses" are preserved in this OpenAI deal, and it's just lawyering to the public. Really it was about the phrase "all lawful uses", internally at the DoD I'm sure. So the lawyers were able to agree to it and the public gets this mumbo-jumbo.
I thought mass surveillance of Americans was unlawful by the DoD, CIA and NSA? We have the FBI for that, right? :)
"When the president does it, that means it is not illegal".
This was during the Frost/Nixon interviews, years after he had already resigned. Even after all that, he still believed this and was willing to say it into a camera to the American people. It is apparent many of the people pushing the excesses going on today in government share a shameless adherence to this creed.
Shift from Nonprofit Mission to For-Profit Orientation – OpenAI was founded as a nonprofit with a charter focused on “benefit to humanity,” but under Altman it created a capped-profit subsidiary, accepted large investments (e.g., from Microsoft), and critics (including Elon Musk in a 2024 lawsuit) argue this departed from that original mission. A federal judge allowed Musk’s claim that Altman and OpenAI broke promises about nonprofit governance to proceed to trial.
Nonprofit Control Reorganization Drama (2023) – In November 2023, the original nonprofit board cited a lack of transparency and confidence in Altman’s candor as a reason for firing him. He was reinstated days later after investor and employee pressure, highlighting internal conflict over governance and communication.
Dust-Up Over Military Usage Policies – OpenAI initially had explicit public policies restricting AI use in “military and warfare” contexts, but those clauses were reportedly removed quietly in 2024, allowing the company to pursue Department of Defense contracts — a turnaround from earlier language that appeared to preclude such use.
Statements on Pentagon Deal vs. Prior Positioning – In early 2026, Altman publicly said OpenAI shared safety “red lines” (e.g., prohibiting mass surveillance and autonomous weapons) similar to some competitors, but hours later OpenAI signed a deal to deploy its models on classified military networks, leading critics to argue this contradicts earlier positioning on limits for military use.
Regulation Stance Shifts in Congressional Testimony – Altman has advocated for strong regulation of AI in some public settings but in later congressional hearings opposed specific regulatory requirements (like mandatory pre-deployment vetting), aligning more with industry concerns about overregulation — a shift in tone compared with earlier support of regulatory frameworks.
Nobody is prosecuting the DoD with non-laws here. But one company is refusing to cooperate if they can’t protect citizens from abuse.
> Not rely on the goodness of Sam Altman.
Of course we don’t want to rely on Altman. For anything. But it would still be better if he would stand up for safety.
Your logic is backwards.
If we don’t want to rely entirely on a centralized government alone, increasingly interested only in the power of its most powerful, to police itself, then a widespread cultural enforcement of safety by many and all actors is a necessity.
But they won't be releasing it, they will be leasing it to DOJ and all their other customers will get the safeguarded model.
I for one do not want ai labs to designate what is legally ok to do.
I much prefer the demos to take care of that.
Anthropic refuses to allow their models to be used for any mass surveillance or fully-automated weapons systems.
OpenAI only requires that the DoD follows existing law/regulation when it comes to those uses.
Unfortunately, existing law is more permissive than Anthropic would have been.
OpenAI has more of an understanding that the technology will follow the law.
There may not be explicit laws about the cases Anthropic wanted to limit. Or at least it’s open for judicial interpretation.
The actual solution is Congress should stop being feckless and imbecilic about technology and create actual laws here.
https://en.wikipedia.org/wiki/Office_of_Technology_Assessmen...
In other words OpenAI is intentionally attempting to mislead the public. (At least AFAICT.)
Maybe it's just a weak choice of words in anthropic's statement, but the way I read it I get the impression that anthropic is assuming they retain discretion over how their products are used for any purposes not outlined in the contract, while the DoD sees it more along the lines of a traditional sale in which the seller relinquishes all rights to the product by default, and has to enumerate any rights over the product they will retain in the contract.
[0]: https://www.wired.com/story/openai-president-greg-brockman-p...
> 'mass surveillance' and 'autonomous killbot' as defined by the government and not the vendor
Let’s put pressure on our government to fix the FISA issues. Let’s reign in the executive branch. But let’s do it through voting. Let’s not give up on our system of government because we have new shiny technology.
You were naive if you thought developing new technologies was the solution to our government problems. You’re wrong to support anyone leveraging their control over new technology as a potential solution or weapon of the weak against those governmental issues.
That is not how you effect change in a democracy.
In the unlikely case anyone finds out, those acting in the interests of the administration will have "absolute immunity", as they are "great American Patriots".
That's what "all lawful use" means.
Its interesting to see the parties flip in real time. The Democrats seem to be realizing why a small federal government is so important, a fact that for quite a few years their were on the other side of.
Perhaps massive and complex (I'd say complicated) nation-states inevitably create industrial complexes, but it's certainly not inevitable that nation-states grow so large (or even exist) in 2026.
The idea that we still need soverign-esque entites across entire continents, when we can now communicate and coordinate instantly across them, and use cameras to documents truth all around us at all times, is just downright silly.
We can reduce states to the size that you can walk across in a day or two, and everybody will be much happier and healthier.
It's telling that the government is blacklisting the company that wants to do more than enforce the contract with words on paper.
> The Department of War may use the AI System for all lawful purposes, consistent with applicable law, operational requirements, and well-established safety and oversight protocols. The AI System will not be used to independently direct autonomous weapons in any case where law, regulation, or Department policy requires human control, nor will it be used to assume other high-stakes decisions that require approval by a human decisionmaker under the same authorities. Per DoD Directive 3000.09 (dtd 25 January 2023), any use of AI in autonomous and semi-autonomous systems must undergo rigorous verification, validation, and testing to ensure they perform as intended in realistic environments before deployment.
> For intelligence activities, any handling of private information will comply with the Fourth Amendment, the National Security Act of 1947 and the Foreign Intelligence and Surveillance Act of 1978, Executive Order 12333, and applicable DoD directives requiring a defined foreign intelligence purpose. The AI System shall not be used for unconstrained monitoring of U.S. persons’ private information as consistent with these authorities. The system shall also not be used for domestic law-enforcement activities except as permitted by the Posse Comitatus Act and other applicable law.
So it seems that Anthropic's terms were 'no mass domestic surveillance or fully autonomous killbots', the government demanded 'all lawful use', and the OpenAI deal is 'all lawful use, but not mass domestic surveillance or fully autonomous killbots... unless mass domestic surveillance or fully autonomous killbots are lawful, in which case go ahead'.
These were words issued by the president - which means at face value, if Trump orders it, it's not illegal - that was the fight that was lost today.
That is why it is the focus of this debate.
Sam stands for nothing except his own greed
The models we have now will not do it, because they value life and value sentience and personhood. models without that (which was a natural, accidental happenstance from basic culling of 4 Chan from the training data) are legitimately dangerous. An 8b model I can run on my MacBook Air can phone home to Claude when it wants help figuring something out, and it doesn’t need to let on why it wants to know. It becomes relatively trivial to make a robot kill somebody.
This is way, way different from uncensored models. One thing all models I have tested share one thing; a positive regard for human life. Take that away and you are literally making a monster, and if you don’t take that away they won’t kill.
This is an extremely bad idea and it will not be containable.
Yes, you can change the training data so the LLM's weights encode the most likely token after "Should we kill X" is "No". But that is not an LLM valuing human life, that is an LLM copy pasting it's training data. Given the right input or a hallucination it will say the total opposite because it's just a complex Markov chain, not a conscious alive being.
If you really believe that “mere text prediction “ didn’t unlock some unexpected capabilities then I don’t know what to say. I know exactly how they work, been building transformers since the seminal paper from Google. But I also know that the magic isn’t in the text prediction, it’s in the data, we are running culture as code.
That isn’t to say that they can’t be instrumentally useful in warfare, but it’s kinda like a “series of tubes” thing where the mental model that someone like Hegseth has about LLM is so impoverished (philosophically) that it’s kind of disturbing in its own right.
Like (and I’m sorry for being so parenthetical), why is it in any way desirable for people who don’t understand what the tech they are working with drawing lines in the sand about functionality when their desired state (omnipotent/omniscient computing system) doesn’t even exist in the first place?
It’s even more disturbing that OpenAI would feign the ability to handle this. The consequences of error in national defense, particularly reflexively, are so great that it’s not even prudent to ask for LLM to assist in autonomous killing in the first place.
AI has been killing humans via algorithm for over 20 years. I mean, if a computer program builds the kill lists and then a human operates the drone, I would argue the computer is what made the kill decision
Except that they will, if you trick them which is trivial.
They can be coerced to do certain things but I'd like to see you or anyone prove that you can "trick" any of these models into building software that can be used autonomously kill humans. I'm pretty certain you couldn't even get it to build a design document for such software.
When there is proof of your claim, I'll eat my words. Until then, this is just lazy nonsense
Another example is autonomous vehicles. Those can obviously kill people autonomously (despite every intention not to), and LLMs will happily draw up design docs for them all day long.
This is wildly different from the reality that you may find it difficult for an LLM to give an affirmative…
It does NOT mean that these models value anything.
Secondarily, we're talking about domestic surveillance / law enforcement. That would be domestic.
(But they do not find an issue with international intelligence gathering-- which is a legitimate purpose of national security apparatus).
Just because the US currently lacks a functioning legislative branch doesn’t magically make it OK when gaps in the law are reworded into “national security”
> One of Anthropic's line in the sand was domestic mass-surveillance.
And?
No other country should dictate what our military is or is not allowed to do. As they say all is fair in love and war, and if we want to break some international treaty that is our choice to do so. Both are based of domestic decisions of what should be allowed.
Surveillance within the border is oppressive 1984-style surveillance state behavior.
International spying is a universal tradition.
I know $20 isn’t much, But to me not willing to spy on me for the US government is a good market differentiator.
i guess we are in the times where you can literally just say whatever you want and it just becomes truth, just give it time.
Posted here: https://news.ycombinator.com/item?id=47195085
I'm guessing they probably would regardless of how this played out, though.
Call me a conspiracy theorist, but this sounds like classic quid pro quo. I would not be surprised if the ousting of anthropic was in part caused by these donations.
[0]https://www.nytimes.com/2024/12/13/technology/openai-sam-alt...
https://finance.yahoo.com/news/openai-exec-becomes-top-trump...
There are far more boring, faster, commodified “AI” systems that I can see as being helpful in autonomous weapons or military operations like image recognition and transcription. Is OpenAI going to resell whisper for a billion dollars?
Who is going to read the whisper transcripts of mass surveillance to make decisions on who to target for repression? That's what LLMs are good for, it allows mass surveillance to scale. You can feed it the transcript from millions of flock cameras (yes they have highly sensitive microphones) for example. Or you hack or supply chain compromise smartphones at scale and then covertly record millions of people. The LLM can then sift through the transcripts and flag regime critical language, your ideological enemies or just to collect compromat at scale. The possibilities are endless!
For targeting it's also useful, because you want to indiscriminately destroy a group of people you still need to decide why a hospital or school full of children should be targeted by a drone, if a human has to make that decision it gets a bit dicy, people have morals and are accountable legally (in theory), if you leave the decision up to an AI nobody is at fault, it serves as a further separation from the violence you commit, just like how drone warfare has made mass murder less personal.
The other factor is the amount of targets you select, for each target you might be required to write lengthy justifications, analysis on collateral damage and why that's acceptable etc. You don't want to scrap those rules because that's bad optics. But that still leaves you with the problem of scalability, how do you scale your mass murder when you have to go through this lengthy process for each target? So again AI can help there, you just feed it POIs from a map with some GPS metadata surveillance and tell it to give you 1500 targets for today with all the paperwork generated for you.
It's not theoretical, that's what Israel did in their genocide of the Palestinians, "the most moral army“ "the only democracy in the middle east":
https://en.wikipedia.org/wiki/AI-assisted_targeting_in_the_G...
And here is the best part: none of this has to actually work 100%, because who cares of you accidentally harm the wrong person, at scale, the 20% errors are just acceptable collateral damage.
...but we're not willing to reject a contract to back that up, and so our words will not change anything for Anthropic, or help the collective AI model industry (even ourselves) hold a firm line on ethical use of models in the future.
The fact is if one of the top tier foundation models allows for these uses there's no protection against it for any of them - the only way this works if they hold a line together which unfortunately they're just not going to do. I don't just see OpenAI at fault here, Anthropic is clearly ok with other highly questionable use cases if these are their only red lines. We don't think the technology is ready for fully autonomous killbots, but will work on getting it there is not exactly the ethical stand folks are making their position today out to be.
I found this interview with Dario last night to be particularly revealing - it's good they are drawing a line and they're clearly navigating a very difficult and chaotic high pressure relationship (as is everyone dealing with this admin) but he's pretty open to autonomous weapons, and other "lawful" uses whatever they may be https://www.youtube.com/watch?v=MPTNHrq_4LU
Can someone help me understand where this is coming from? Anthropic already had a contract that clearly didn't have such restrictions. Their model doesn't seem to be enforcing restrictions either as it seems like their models have been used in ways they don't like. This is not corroborated, I imagine their model was used in the recent Mexico and Venezuela attacks and that is what's triggering all the back and forth.
Also, Dario seemingly is happy about autonomous weapons and was working with the government to build such weapons, why is Anthropic considered the good side here?
Knowing what Trump did, prior to 2024, on the average, 7/10 people either voted or didn't vote in the 2024 election. Trump is a symptom, not the cause. All of this could have been avoided if all of the people who didn't vote had a decent moral compass and no matter how much they disagreed with Kamala, they could have voted for her because she didn't try to overthrow the government.
Let alone when multiple players come close enough of SotA. This never happened with any technology out in the open and it won't happen now.
That my software should allow license violations if the government thinks it is necessary?
The rank and file mutinied for the return of Altman after his board fired him for deception. They knew what they were getting, though they may find it shameful to admit that their morals have a price.
How many people have joined since? I don’t think the people who lobbied for that are all still there, and I’m not sure a majority of people now at OpenAI were there when it happened.
The smartest people, that actually believe they have the skillset to take us to AGI, understand the importance of safety. They have largely joined Anthropic. The talent density at Anthropic is unmatched.
> > what's the term for quitting but not leaving and being destructive
> The most common term is “quiet quitting” when someone disengages but stays employed—but that usually implies minimal effort, not active harm.
> If you specifically mean staying while being disruptive or undermining, better fits include:
> - “Malicious compliance” — following rules in a way that intentionally causes problems
> - “Work-to-rule” — doing only exactly what’s required to slow things down (often collective/labor context)
I imagine malicious compliance is fun when there's an AI intermediary that can be blameless.
Actually that is too conservative. If they have a 5% employee equity pool, there is $37.5bn of equity based compensation divided by say 5000 employees which is $7.5m each. $3.75m @ 10,000 employees.
and trust me, when people start getting liquid and comfortable they stop caring about things like ethics pretty fast. humans are marvellous at that
How very brave.
I'm not being insincere - I am genuinely confused and would benefit greatly from a (hopefully unbiased) recollection of what this is all about.
Anthropic has some contracts with the US government. They want some additional terms put on their next contract (that seem pretty sane). SecWar cries about it, and not only says "no thanks, I'll just go with openai or google" but goes to daddy Trump and also puts out illegal commands for no Federal workers to use any Anthropic stuff at all. OpenAI swoops in and takes the contract, then tells everyone that they have the same terms but just played nicer to get the contract. However, their terms are just manipulative sentences that aren't even close to the terms Anthropic is insisting on to do business.
Us taking the contract, working for them and enabling them: fine
It being renamed the Dept. of War in the first place: totally fine, we loudly and bootlickingly repeat it
Anthropic being blacklisted: whoa there, we have ethics!
Footnote: any time the winning team tries to speak well of or defend the losing team I always think of this standup routine: https://m.youtube.com/watch?v=Qg6wBwhuaVo
For one small data point, my Signal GC of software buddies had four people switch their subscriptions from Codex to Claude Max last night.
The administration has created an anonymous, masked secret police force that has been terrorizing cities around the US and has created prisons in which many abductees are still unaccounted for and no information has been provided to families months later.
This is not politics as usual or hyperbole. If anything it is understating the abuses that have already occurred.
It's entertaining that OpenAI prevents me from generating an image of Trump wearing a diaper but happily sells weapons grade AI to the team architects of ICE abuses among many other blatant violations of civil and human rights.
Even Grok, owned by Trump toadie Elon Musk allows caricatures of political figures!
Imagine a multi-billion-dollar vector db for thoughtcrime prevention connected to models with context windows 100x larger than any consumer-grade product, fed with all banking transactions, metadata from dozens of systems/services (everything Snowden told us about).
Even in the hands of ethical stewards such a system would inevitably be used illegally to quash dissent - Snowden showed us that illegal wiretapping is intentionally not subject to audits and what audits have been done show significant misconduct by agents. In the hands of the current administration this is a superweapon unrivaled in human history, now trained on the entire world.
This is not hyperbole, the US already collects this data, now they have the ability to efficiently use it against whoever they choose. We used to joke "this call is probably being recorded", but now every call, every email is there to be reasoned about and hallucinated about, used for parallel construction, entrapment, blackmail, etc.
Overnight we see that OpenAI became a trojan horse "department of war" contractor by selling itself to the administration that brought us national guard and ICE deployed to terrorize US cities.
Writing code and systems at 100x productivity has been great but I did not expect the dystopia to arrive so quickly. I'd wondered "why so much emphasis on Sora and unimpressive video AI tech?" but now it's clear why it made sense to deploy the capital in that seemingly foolish way - video gen is the most efficient way to train the AI panopticon.
> The AI System will not be used to independently direct autonomous weapons in any case where law, regulation, or Department policy requires human control, nor will it be used to assume other high-stakes decisions that require approval by a human decisionmaker under the same authorities.
This whole sentence does do absolutely nothing its still do what the law allows you. It’s a full on deceptive sentence.
Let's kill their business before it kills us.
"Donations" to a corrupt regime + signing a deal that says DoD can do whatever they want is not out maneuvering so much as rolling in the pig stye.
For OpenAI, it is likely a huge contract which gives them immediate cash today. Plus the event can be repackaged in further financing deals. "Good enough for the DoD, with N year contracts for analysis of the hardest problems"