Suppose a company calls themselves The Doomsday Device Company. They make and sell excellent-quality doomsday devices. They regularly go online to proclaim that their doomsday devices are the best and most powerful, and also that doomsday devices are dangerous and should be regulated.

The Doomsday Device Company then says they have the world's best doomsday device. (They don't, but they claim they do.)

The US Government hates the Doosmday Device Company for various political reasons, but also has a vested interest in there not being a massive proliferation of doomsday devices.

The Doosmday Device company spends a great deal of time and money telling everyone: "Our doomsday device is the most doomy of all time!" (though it probably isn't) and "Everyone can use it!" (for a lot of money)

It is completely logical, then, for the US Government to say: No, everyone cannot use your doomsday device, because doomsday is bad. (While also meaning: Only we should be able to use it, and you shouldn't be able to tell us how.)

If you do not want to be in the business of having your doomsday devices shut down by the government, well, it would help if you didn't so loudly and aggressively proclaim how doomy they are. It doesn't matter how trustworthy you claim to be, given that your business is making evil doosmday devices. You still won't be trusted!

Companies/countries/people are paying off the government in all sorts of various ways (crypto, gifts, bogus settlements, planes, inaugurations, ballrooms). The companies that pay off the government get big fat contracts and merger agreements, and the ones that don't get increased scrutiny, lawsuits and threats.

OpenAI and SpaceX are friends of the administration, and Anthropic is (politically at least), not friends with the administration.

Could this penalty be a rational and reasonable reaction to the new model? Perhaps. Or maybe it is just a made up excuse to do what the government wants to do, which is punish its political enemies. It wouldn't be the first, second, third or 10th time that has happened so far in this administration.

I.e. OpenAI just went full evil corpo mode and went all in on the Leading the Future PAC [1] to try and prevent any kind of regulation.

I feel like there is a reasonable path where they might agree with OP that the government has "mostly gone insane" but also think that US getting its act together and leading the way on sane regulation will be key to getting to a good outcome with AI.

[1]. https://en.wikipedia.org/wiki/Leading_the_Future

So I genuinely think that Amodei thought here that he was building a moat - set a very high bar for safety at exactly the line Anthropic but nobody else meets, and then declare anything less to be too unsafe to be allowed. That would put a permanent halt to open models, Chinese models and throw a significant barrier in front of competitors - if OpenAI is about to release something competitive with Mythos, they would have to immediately double back and implement at least equivalent safeguards. It might cost them months at the most critical juncture in Anthropic's history, when they are filing for IPO.

Having said this, I am sure they calculated in the possibility of their own model being restricted. They probably still see it as a win because it acts as a strong endorsement of them as the market leader and the model as the most powerful available model. So I think both things are true, but we are in the "plan B" scenario now rather than "plan A".

https://www.anthropic.com/institute/recursive-self-improveme...

> We believe it would be good for the world to have the option to slow or temporarily pause frontier AI development to enable societal structures and alignment research to keep up with the advance of the technology.

In their subsequent post this week responding to the announcement of the export ban, Anthropic wrote:

> If this standard was applied across the industry, we believe it would essentially halt all new model deployments for all frontier model providers.

Which is what they said would be good.

Didn't Anthropic say that the same jailbreak is possible with GPT 5.5?

> I believe there are: they are called “courts”. Dario is as free as the rest of us are to file a lawsuit and go in front of a judge and tell the judge that he is the victim of political favoritism or an arbitrary decision. That is, in fact, one of the primary purposes of the legal system.

This isn't realistic here. Yes, there's a system in place, but at the speed of these iterations/deployments, filing a lawsuit that will take months/years to resolve isn't a practical path forward.

Having the US control export your flagship model is the ultimate stamp approval in the AI race. All headliners are american, but one us too poweful to be made available. It just reads like if someone was riding an IPO.

It is also, paradoxically, a wink to the 90's. But we're not in the 90's, and the cat is out of the box, and in 6-12 months everyone else will be at this tier. This is, clearly, an attempt to boost a model that isnt that revolutionary. I used Fable, and for my work, its mostly a waste of tokens. It seems a bit better than Opus 4.8 - but 4.8 the past week(s) has actually been top knotch; so lets make it a "myth" and have secops tell stories about it, so everyone will pay when the time comes; will you, ceo/cto, allow your company to fall behind? Of course not. You will pay. For modest results, apparently, but the hype is there.

Most of the people complaining about Anthropic's behavior, while simultaneously avoiding the argument at heart about whether AI regulation is good, remind me of the "we should improve society" meme:

https://iea.org.uk/wp-content/uploads/2021/12/mister-gotcha-...

> I think all of this was extremely irresponsible of them, and I feel a good amount of schadenfreude that the leopard ate their face first.

Can't say I disagree. Hope this costs them many, many billions.

They will either play ball with the US government on the US government's terms or they will be replaced or destroyed.

It is a misconception for them to believe they can dictate the terms of this technology.

In my mind, as recently as February, Anthropic was considered by far “the best” company on the planet, with an insane fanbase and praise left, right, and centre. The story around the DoD contract solidified them in the public eye as “the hero the city deserved.”

Then they fell into a classic monetisation trap. Unable to sustain growth at such a discount, they started nerfing models, removing caching, and doubling costs per token to make any money here.

The lack of transparency in that process cost them public perception. By early May, all the charm of the magical, ethical super-company was gone. The entire campaign around the Mythos release, intentional or not, landed on top of that new narrative and didn’t play well for them.

What is interesting to me here is the realisation that a good chunk of the hate the company received came simply because of their most recent hostile, for-profit actions. Had it happened in March, HN and the public reaction would have been vastly different. It took them just two months of “bad actions” to ruin quite a good margin of the public praise they so desperately needed now.

> We’re proposing stronger regulation of the technology. We’re proposing giving the government the ability to, again in a narrow way, block deployment of unsafe technology.

Anthropic CEO, last week.

https://abc7.com/post/anthropic-ceo-dario-amodei-calls-stron...

If you sold everyone on the idea of "safety is paramount, we urge everyone not to rush into development here" then certainly becomes hard to believe a blanket "we figured out safety, come play with our toys for 10x cost" when stuff is less than a month apart in your news page.

Under such conditions we would be looking at Amazon's actions through a much more benevolent lens.

Not saying it has been, but it certainly crossed my mind as something worth doing regardless.

They've positioned their company as 'We're the serious AI company that understands safety, while others underestimate the risks.' That strategy itself is understandable. They're not like OpenAI, which carved out the pioneer position in LLMs, nor do they have a trustworthy brand like Google (Gemini isn't trustworthy, but still). So branding around 'responsibility' made sense.

The problem is that they pushed that narrative with the Trump administration. Without considering that LLM strategies need to change depending on the political context, they just input the same prompt into a different context and got bad results.

The Trump administration's stance emphasizes external enemies. I guess they didn't know what would happen if they started talking about military weapons in that environment.

We East Asians know authoritarian regimes 'very' well. So I guess people from the US, a country with so much freedom that they naturally lie flat on the ground, just didn't understand the difference.

If they had advocated for AI freedom and free expression, many people might have helped them, like in the PGP situation in cryptography. But instead, they got caught up in their own claims.

If you emphasize how dangerous AI is under an administration like Trump's that stresses external enemies, of course the government will say, 'Then let us manage it.' And the moment Anthropic says, 'Why just us?' it just looks ridiculous. They're the ones who went on about how dangerous it is, and now they're acting victimized for being treated as a dangerous entity.

To be even more honest, Amodei's style of communication sometimes looks like a morality superiority hustle.

They speak in a tone of 'We're not just a money focused company, we care about humanity,' but isn't Anthropic still a company that takes investments, sells models, rides the cloud, and tries to win government contracts? So it ends up looking like they use regulatory discourse as a shield and marketing when it benefits them, but complain 'it's not fair' when it works against them.

Personally, I think Anthropic needs to hire a Korean person as their marketing lead. We Koreans know very well how to behave under authoritarian governments. If you need a marketing person, feel free to contact me. I'll prepare my resume

None of the large language model providers have a very defensible product moat yet and if the models themselves can reveal research fundamentals their position would become extremely precarious.

It would be very tempting to hide behind a national security excuse to try and preserve the research moat.

But is this the type of regulation Anthropic has been asking for? Not at all.

This is a failure of Anthropic's politicking, and a warning that they need to be more careful with their communication in the future. If they truly want constructive regulations because of their fears about AI, they will need to repair their relationship with the administration, and it is still unclear to me how they plan to do that.

"The government should have the power to block or deter deployment of the model if it is determined, in light of third-party assessment, to present unacceptable risks. This power must be scoped to the above four specific risks and there must be protective measures against political favoritism or arbitrary decisions."

The article uses this as a gotcha, but I think there is nuance here. Amazon is a competitor, not an officially appointed third-party that can assess anything. And while Amodei says government oversight is needed, I think it's fair to say we would all expect something more rigorous and neutral than getting a phone call on a Friday based on vibes.

The article is writing as if Amazon did a complex analysis and then reported it.

But the latest reporting id read was it was not a jailbreak, and reported by the ceo (not the old technical CEO btw, the new bizdev guy)

That framework is, of course, what Amodei did ask for, but he mistakenly thought he'd have a seat at a table populated by rational actors. Even after the Trump administration explicitly told him otherwise when they declared his whole company to be a national security threat.

So what happened is all Amodei's fault. It's possible that the Anthropic board will decide that this particular unforced error is his last one. In fact, given that Amazon is apparently the prime mover behind this whole train wreck, I'd almost bet on it.

Weapons.

Know what’s not? Words. Text. Tool calls.

Dario has his name on the OpenAI claims years ago that “gpt-2 is too powerful to release”, so he’s been pulling this base-of-the-brain-stem crap for years at this point. When you unpack what they’re saying, it boils down to “The Chinese and everyone working in the open are really scary and we should stop their progress”. Remind me again whose shoulders you sit on?

Even more disgusting, he did it with an unnecessarily long essay. You know that colleague who puts everything in ChatGPT and you’re forced to read all of it just in case there’s a buried surprise?

Anthropic is doing that to government.

The strategy only makes sense - and the essay is only feasible to write - if you have access to “a country full of sycophants in a datacenter.”

    $ xdg-open fakerake.png
    Claude: regulate me
    USA: YOU ARE BEING REGULATED
    Claude: oh my god

As someone that doesn’t have a dog in this race, I feel like anthropic has been very consistent with their moral stance. First, they denied the Department of war to use their AI to conduct military operations and throughout all this, Anthropic has been the one to neuter their model and make sure that it’s not able to do a lot of things that might can be destructive. So them saying that there should be a pause on new AI and then releasing this new product makes me inclined to believe them. Maybe I’ve drank the koolaid but it seems like Anthropic isn’t inherently “evil.”

>Yes. This assessment was made by Amazon, a frequent and serious government contractor which is generally trusted to handle high-security government, intelligence, and military contractor concerns.

Reads as partially disingenuous. Amazon did not conduct some thoroughly vetted, responsible security audit. Someone gave them examples of a 'jailbreak' and they notified the white house rather quickly. This was nary an official process. Calling it one is ignoring the facts of what happened.

He rather obviously is asking to establish a 3rd party specifically for this task, and to establish guidelines relating to the given concerns, and to establish guidelines for government actions based on the evaluation by the 3rd party.

Anthropic for lying too much and for too long about their capabilities

US Government for taking these marketing stunts for ground truth

EU for freaking out about its Digital Sovereignty over US companies, as if it had it before.

They got even more than what they asked for.

Remember, all AI companies openly claimed to oppose military usage just a few short years ago. Now they all have government contracts that allow the government to use them "lawfully", while also being able to decide that anything they do with them is lawful. Anthropic is the only one who required clauses against killbots and domestic mass surveillance.

Anthropic never asked for arbitrary or opaque shutdowns. They asked for clearly defined regulations to apply equally (which would've helped their market position and advantage, coincidentally I'm sure /s), moreso to reduce their own risk and liability.

1. "Dario is known for writing about regulation and the direction of AI as an industry and Anthropic in particular, and what he says is taken very seriously and is considered a definitive statement of the company’s position." This is patently ridiculous. A CEO's blog post is not an official company statement or any sort of binding agreement.

2. "Are there protective measures against political favoritism or arbitrary decisions? I believe there are: they are called “courts”." This is so stupid. Of course Anthropic will take this to court (if it's not rescinded before then), and the government's ham-fisted "regulation" will almost certainly be overturned. And it doesn't matter! An unjust action that is overturned by the legal system does not magically become just.

3. "Is This Politically Motivated or Arbitrary? Probably at least somewhat." If the best you can muster here is "probably at least somewhat", then your head is in the sand. It clearly politically motivated, and clearly arbitrary. Perhaps a different government would receive the benefit of the doubt here, but not this one.

4. "“The government” or “society” is meant to deal with all of those things. Well, now the government is — the actual government that really exists, and not an imagined one that only does good things and never does bad things." So that's it? We just throw up our hands and say that this is natural, that it couldn't go any other way? That Anthropic was "asking for it", and it's their fault when the government lashes out?

If the government wants to regulate AI, either Congress needs to pass a law, or the Executive needs to furnish a reasonable explanation for their actions. We do not live in a fascist country. There is separation between the government and private industry. The government does not have the power to arbitrarily regulate private enterprise. I am truly baffled by the inability for people to see this as it is -- a blatant, and foolish, attempt at posturing and political intimidation. It's part of a clear pattern of behavior by this administration, and should be interpreted as such.

The government can't apply export controls based on a control that does not exist. Creating one for model inference, if at all possible, would take 3-6 months at a minimum and it even includes a public comment process. That control is not cited anywhere because it does not exist.

The president can invoke emergency powers but that requires pointing to a specific foreign threat, notifying congress formally, posting on the national register, and it only lasts six months unless congress votes to renew it.

Given how easy it would have been for them to fight this, we can only conclude this was either outright designed or incredibly convenient for Anthropic.

Given their stated goal of pulling Fable by June 22nd, it seems likely they underestimated the amount of compute they would need or, even if they had perfectly estimated it, pivoting so that "the government shut it down because it's so powerful" on June 12nd is a better story than "we shut it down because we lack the compute" on the 22nd. This is especially true because the net new revenue from Fable is just from new signups between the two dates, which is likely smaller each day elapsed since the launch.