> Pillar 2: AI safety is a universal practice backstopped by law. AI developers everywhere follow safe development and deployment practices, sustained by industry norms and enforced by regulation.
They so, so badly want to criminalize local models.
edited:
> Interpretability isn’t reliable enough to fully understand model intentions: Being able to read and understand a model’s internal reasoning reduces our reliance on what the models explicitly tell us about why they behaved in a certain way. This could be a useful tool for understanding models’ “thought processes” when they take unintended actions. But interpretability methods are not yet
dependable or broad enough to be relied on in this way. As a result, we may miss our assessment of high-risk behaviors, or be able to investigate flawed reasoning only after an
incident has occurred.
Just say you have no idea how or why they do what they do. Just the other day I saw it make an obnoxious, huge, irrelevant comment in a block of code - many lines more than the code change (which by the way, I have tons of instructions and hooks intended to stop this, it's maybe 95% reliable at best) - and told it quite plainly, "remove that comment."
What it did instead was remove most of the comment except one line, which was also irrelevant. When I noticed, I asked it why it did that when I explicitly stated otherwise and if it misunderstood what I meant. It happily chimed "No, I did not misunderstand your intent, I just thought this was better."
I can't trust that was the real reason so clearly I can't trust that whatever training causes this behavior results in a reliable tool.
I didn't want to keep editing my comment so I am replying to it, so much of this document makes me angry:
> Before we can align a model with human values, we must consider what those
values should be.
Ok, so you believe that 1) this technology will become so powerful as to surpass humans in pretty much every imaginable way, so much so that you see it as an existential threat to humanity itself, and 2) that you see yourself as a responsible steward to avoiding bad outcomes, but 3), that you need to be the arbiter of what human values should be?
A for profit company with shareholders designing technology they think will take over the world and run most if not all of humanity, is deciding for itself what human values should be doesn't set off any alarm bells?
They have to decide what "human values" their model is going to be adapted towards. Ultimately, they have to decide that, even if they choose an outside source. How else are they supposed to do it?
> Before we can align a model with human values, we must consider what those
values should be. In 2023, we defined our intentions and goals for Claude’s
behavior in a constitution, which we have refined over time. In January 2026,
we published a new version under an open license so others could adapt it.
Seems like they're pointing out a pretty uncontroversial fact: if you're building a model and want it to be aligned with some set of values, one of the first things you need to do is determine the set of values. So are you arguing that (1) Anthropic shouldn't be trying to align models, (2) Anthropic shouldn't be trying to build models at all, (3) the values described in the relevant documents are bad, (4) there should have been a democratic process determining the values that Anthropic would be allowed to try to instill in their models, or something else?
Let’s take it the other way around. If it is so dangerous and needs to be regulated, why is it not dangerous enough to be nationalised like an army? Even in this case, for countries with militia, such as US, why people are trusted to care a gun but would not be trusted to run a model. The fear that local models will soon be sufficient for most workloads is palpable. This is not regulatory capture for competition it is for devaluation of their product.
There are different economists who see things differently, like Nobel Laureates Daron Acemoglu et al, who wrote a paper titled "Building Pro-Worker AI"
Giving the public an interest in the private company gives the public the impression that it's in their benefit if the government doesn't regulate the company.
Like giving turkeys a share in a cranberry sauce company.
Some of the aspects of these pillars need to be fleshed out, this reads like it was slapped together as a last second presentation.
For example, “Help all people, especially those most affected by AI, earn more and feel financially secure.”
How would this work? How does AI do anything but detract from the value of labor? Anything made by AI or through heavy AI involvement can be remade just as easily. It circumvents the moat of expertise and effort.
I wish AI companies would just be honest with the most likely consequences of their products proliferation, instead of these Disney-a** sunshine and rainbow pieces. Then maybe we could have a real conversation.
The pillars read as claimed achievements in this link. But from the charter: "Each pillar describes a condition we want to be true in a world where AI has gone well, and realizing them is a project that encompasses not only Anthropic but also governments, academia, civil society, and the broader AI industry."
The public would benefit the most from an American distillation regime. The faster models become a commodity the better. Then it's truly a public good.
Maybe you're thinking of B Corp? That's just a certification from some group. Public Benefit Corporation, on the other hand, changes directors' fiduciary responsibilities. For example, earlier this year, the Delaware Court of Chancery ruled that PBC directors are not obligated to maximize shareholder value when selling control of their company.
I highly recommend Eric Ries' new book Incorruptible if you want to learn more about this stuff. In there he tells the story of the Vectura Group, a pharma company that was developing treatments for asthma and other respiratory diseases. It was looking at a sale, and it had a couple offers, the one from Phillip Morris slightly edged out the other, and because of fiduciary duty, even though many involved viscerally hated the idea of going to Phillip Morris, as the cause of many of the ailments they sought to cure, they sold to them.
Could you elaborate on your feelings on that? I was reading about BlueSky's PBC status literally yesterday, and am rather interested in what people's perception of such an entity is, vs a hybrid such as Wikipedia/Enterprise Wikimedia.
What a crock of shit. Who actually buys the idea that they are in for anything but money? Pretending to care is almost worse than just owning the fact they are out for profit.
Every single "pillar" they mention is demonstrably false.
They are the only major US AI company who objected to the use of their AI for murder without human oversight. They got canceled by the government for that.
Yeah, I find the HN hate for anthropic over other AI companies somewhat odd. Are you honestly trying to have me believe you trust them less than OpenAI or Meta? It honestly reads like a bunch of employees of the also rans complaining about the current leader in the space.
I don't trust Anthropic, OpenAI, Meta or X. They are all run by some of the least trustworthy people I've ever seen and have all demonstrated, many times over, that they should never be trusted. Open models are the only way to go.
False dichotomy: I don't trust any of them. The current AI frontrunners are primarily distinguished by being the ones who were able to raise the most capital, and they hang out in similar circles to Sam Bankman-Fried.
It's really just the constant fear mongering against open models, the overly messianic tone of their marketing, and the endless hinting that "Claude" is a conscious being fully deserving of the same rights as a human.
Wow, they did one good thing! Let's absolve them of everything else. Saying "they objected to murder!" is not the defense you think it is. And what does it say about the other AI companies? They are all a cancer, and they pretend it's for everyone's benefit, and they pretend we have some say in what they do. "But they didn't murder anyone!" is the bare fucking minimum I expect from every company, and a whole lot more besides.
She probably thought that having Jeffrey Epstein involved in her porn site would be a benefit to humanity. The people involved in Epstein's plots probably also thought they were benefiting humanity (just maybe not the humans alive right now)
And for clarity, this isn't just Dario's wife, she's part of the executive staff.
I was never a fan of Anthropic but I like them less and less as the year wears on. Claude increasingly has begun to moralize and it sometimes sounds like a holier-than-thou NorCal purity person.
I increasingly consider Anthropic's so-called humanistic stance completely hubristic and hypocritical. They are our self-appointed future dictators.
I also hate Claude's moralizing but as a "NorCal purity person" I can say unequivocally that I don't include murdering over 120 little girls in my morals.
> Pillar 2: AI safety is a universal practice backstopped by law. AI developers everywhere follow safe development and deployment practices, sustained by industry norms and enforced by regulation.
They so, so badly want to criminalize local models.
edited:
> Interpretability isn’t reliable enough to fully understand model intentions: Being able to read and understand a model’s internal reasoning reduces our reliance on what the models explicitly tell us about why they behaved in a certain way. This could be a useful tool for understanding models’ “thought processes” when they take unintended actions. But interpretability methods are not yet dependable or broad enough to be relied on in this way. As a result, we may miss our assessment of high-risk behaviors, or be able to investigate flawed reasoning only after an incident has occurred.
Just say you have no idea how or why they do what they do. Just the other day I saw it make an obnoxious, huge, irrelevant comment in a block of code - many lines more than the code change (which by the way, I have tons of instructions and hooks intended to stop this, it's maybe 95% reliable at best) - and told it quite plainly, "remove that comment."
What it did instead was remove most of the comment except one line, which was also irrelevant. When I noticed, I asked it why it did that when I explicitly stated otherwise and if it misunderstood what I meant. It happily chimed "No, I did not misunderstand your intent, I just thought this was better."
I can't trust that was the real reason so clearly I can't trust that whatever training causes this behavior results in a reliable tool.
The hypocrisy of the second quote is a bit on the nose. If the internal reasoning is so valueable, why do they increasingly hide it from the user?
I didn't want to keep editing my comment so I am replying to it, so much of this document makes me angry:
> Before we can align a model with human values, we must consider what those values should be.
Ok, so you believe that 1) this technology will become so powerful as to surpass humans in pretty much every imaginable way, so much so that you see it as an existential threat to humanity itself, and 2) that you see yourself as a responsible steward to avoiding bad outcomes, but 3), that you need to be the arbiter of what human values should be?
How fucking arrogant, pardon my language. My god.
I don't understand. Human values are not universal, don't they have to be the arbiter? Isn't that all that statement is saying?
A for profit company with shareholders designing technology they think will take over the world and run most if not all of humanity, is deciding for itself what human values should be doesn't set off any alarm bells?
They have to decide what "human values" their model is going to be adapted towards. Ultimately, they have to decide that, even if they choose an outside source. How else are they supposed to do it?
Why on earth should they be the arbitrer? Do people in SF understand how much the rest of the world despises them with this attitude?
Are you sure the "we" there is Anthropic, or is it humanity as a whole?
Why would a for profit company beholden to shareholders listen to anything but their own financial interests? Has this ever happened in recent memory?
Substitute anthropic/claude with exxon mobil/oil and see how crazy this sounds.
No one group or individual should be saddled with deciding what the “correct” human values should be. That’s literally what religion does.
OK, you got me to look at the context:
> Before we can align a model with human values, we must consider what those values should be. In 2023, we defined our intentions and goals for Claude’s behavior in a constitution, which we have refined over time. In January 2026, we published a new version under an open license so others could adapt it.
Seems like they're pointing out a pretty uncontroversial fact: if you're building a model and want it to be aligned with some set of values, one of the first things you need to do is determine the set of values. So are you arguing that (1) Anthropic shouldn't be trying to align models, (2) Anthropic shouldn't be trying to build models at all, (3) the values described in the relevant documents are bad, (4) there should have been a democratic process determining the values that Anthropic would be allowed to try to instill in their models, or something else?
They only care about regulation when it can be used to snuff out competition.
Yes, this is textbook regulatory capture.
https://en.wikipedia.org/wiki/Regulatory_capture
"this is textbook regulatory capture" is a common sentiment in recent ai discourse, often disputed by economists. cf alex tabarrok: https://marginalrevolution.com/marginalrevolution/2026/09/wh...
Let’s take it the other way around. If it is so dangerous and needs to be regulated, why is it not dangerous enough to be nationalised like an army? Even in this case, for countries with militia, such as US, why people are trusted to care a gun but would not be trusted to run a model. The fear that local models will soon be sufficient for most workloads is palpable. This is not regulatory capture for competition it is for devaluation of their product.
There are different economists who see things differently, like Nobel Laureates Daron Acemoglu et al, who wrote a paper titled "Building Pro-Worker AI"
https://www.nber.org/papers/w34854
Alex Tabarrok is certainly a name one can throw into the conversation
I would take the opinions of a right wing blogger with a grain of salt.
Giving the public an interest in the private company gives the public the impression that it's in their benefit if the government doesn't regulate the company.
Like giving turkeys a share in a cranberry sauce company.
Gobble gobble
> AI’s benefits are shared by everyone
Okay so where's my free access to Claude Code
Some of the aspects of these pillars need to be fleshed out, this reads like it was slapped together as a last second presentation.
For example, “Help all people, especially those most affected by AI, earn more and feel financially secure.”
How would this work? How does AI do anything but detract from the value of labor? Anything made by AI or through heavy AI involvement can be remade just as easily. It circumvents the moat of expertise and effort.
I wish AI companies would just be honest with the most likely consequences of their products proliferation, instead of these Disney-a** sunshine and rainbow pieces. Then maybe we could have a real conversation.
They must be consulting Kim Yong Un or similar for the daily propaganda. This is unbelievable.
Maybe organize a "spontaneous" demonstration in favor of AI with 10 year olds wearing scarves and waving Anthropic flags.
The pillars read as claimed achievements in this link. But from the charter: "Each pillar describes a condition we want to be true in a world where AI has gone well, and realizing them is a project that encompasses not only Anthropic but also governments, academia, civil society, and the broader AI industry."
So these are aspirational.
Pillar 6: war is peace
Pillar 7: freedom is slavery
The public would benefit the most from an American distillation regime. The faster models become a commodity the better. Then it's truly a public good.
I expect Anthropic's status as a public benefit corporation to matter exactly as much as OpenAI's status as a nonprofit did
I love how the "public benefit" is just them trying to force everyone to do AI their way
Public benefit is the same as openness of open AI.
PBC is a scam word. It changes nothing about the corporations responsibility.
Maybe you're thinking of B Corp? That's just a certification from some group. Public Benefit Corporation, on the other hand, changes directors' fiduciary responsibilities. For example, earlier this year, the Delaware Court of Chancery ruled that PBC directors are not obligated to maximize shareholder value when selling control of their company.
I highly recommend Eric Ries' new book Incorruptible if you want to learn more about this stuff. In there he tells the story of the Vectura Group, a pharma company that was developing treatments for asthma and other respiratory diseases. It was looking at a sale, and it had a couple offers, the one from Phillip Morris slightly edged out the other, and because of fiduciary duty, even though many involved viscerally hated the idea of going to Phillip Morris, as the cause of many of the ailments they sought to cure, they sold to them.
>I highly recommend Eric Ries' new book Incorruptible if you want to learn more about this stuff.
Looks like an interesting read, thanks for the recommendation!
Could you elaborate on your feelings on that? I was reading about BlueSky's PBC status literally yesterday, and am rather interested in what people's perception of such an entity is, vs a hybrid such as Wikipedia/Enterprise Wikimedia.
I think it would be a public benefit if they stopped aggregating all their accumulated debt on the public.
What a crock of shit. Who actually buys the idea that they are in for anything but money? Pretending to care is almost worse than just owning the fact they are out for profit.
Every single "pillar" they mention is demonstrably false.
They are the only major US AI company who objected to the use of their AI for murder without human oversight. They got canceled by the government for that.
Yeah, I find the HN hate for anthropic over other AI companies somewhat odd. Are you honestly trying to have me believe you trust them less than OpenAI or Meta? It honestly reads like a bunch of employees of the also rans complaining about the current leader in the space.
I don't trust Anthropic, OpenAI, Meta or X. They are all run by some of the least trustworthy people I've ever seen and have all demonstrated, many times over, that they should never be trusted. Open models are the only way to go.
I trust none of them. How is that hard to understand? Being marginally better than OpenAI or Facebook is like preferring a solid turd to a liquid one.
False dichotomy: I don't trust any of them. The current AI frontrunners are primarily distinguished by being the ones who were able to raise the most capital, and they hang out in similar circles to Sam Bankman-Fried.
It's really just the constant fear mongering against open models, the overly messianic tone of their marketing, and the endless hinting that "Claude" is a conscious being fully deserving of the same rights as a human.
I think people can't stand anthropic's pseudo humanism, it comes across as lying and fake. Altman and Zucc don't pretend to be heroes
Wow, they did one good thing! Let's absolve them of everything else. Saying "they objected to murder!" is not the defense you think it is. And what does it say about the other AI companies? They are all a cancer, and they pretend it's for everyone's benefit, and they pretend we have some say in what they do. "But they didn't murder anyone!" is the bare fucking minimum I expect from every company, and a whole lot more besides.
Anthropic was involved with the Minab school bombing. They are liars and the other US AI companies are no better.
Unless they are a Public Benefit Corporation, they will eventually go the way of Google and their 'Don't be evil' motto.
Honestly fuck Anthropic, fuck OpenAI and fuck all the LLM providers.
As soon as I can I'm off to self hosting and providing the service to friends and family via a VPN.
These companies are just ingesting public and frequently private information and regurgitating it back as a service.
As soon as I can run inference locally will do so.
Just look at the maths people whose work was stolen.
You want to benefit the public? Open your weights and training data.
Otherwise FO pretending to have a social conscience, you don't.
> Our charter is to responsibly develop and maintain advanced AI for the long-term benefit of humanity.
Seems like a good time to remember that Dario Amodei’s wife pitched Jeffery Epstein on funding her porn site.
https://www.forbes.com/sites/alisondurkee/2026/08/14/who-is-...
She probably thought that having Jeffrey Epstein involved in her porn site would be a benefit to humanity. The people involved in Epstein's plots probably also thought they were benefiting humanity (just maybe not the humans alive right now)
And for clarity, this isn't just Dario's wife, she's part of the executive staff.
I was never a fan of Anthropic but I like them less and less as the year wears on. Claude increasingly has begun to moralize and it sometimes sounds like a holier-than-thou NorCal purity person.
I increasingly consider Anthropic's so-called humanistic stance completely hubristic and hypocritical. They are our self-appointed future dictators.
I also hate Claude's moralizing but as a "NorCal purity person" I can say unequivocally that I don't include murdering over 120 little girls in my morals.
https://theoutpost.ai/news-story/pentagon-investigates-ai-ro...