"Anthropic Is Not a Safe AI Company" – Coalition Challenges Anthropic Safety
111 points by sshah03 3 hours ago | 58 comments

neya 2 hours ago
Anthropic heavily lobbies and invests millions into lobbying. The only company to do so. And their agenda has always been to stand against open source models because they know they can't win that war. It's no co-incidence that Jensen's leaks were timed perfectly right before Anthropic's IPO. Jensen was a huge advocate for open weight models.

They're also far from caring about anyone's privacy and were caught spying on their users so..

reply
strictnein 2 hours ago
> "invests billions into lobbying"

Off by a factor of ~1000:

"The Financial Times reported that Anthropic nearly tripled its lobbying expenditure to $3.53 million."

https://finance.yahoo.com/technology/ai/articles/anthropic-n...

reply
deflator 46 minutes ago
We are now conditioned to think in terms of billions from the tech buildout, but lobbying in the US is a comparative bargain. You can support the billions with a few million in lobbying.
reply
Mistletoe 43 minutes ago
It has to be the best ROI out there honestly.
reply
catlifeonmars 32 minutes ago
Can confirm firsthand
reply
therein 32 minutes ago
It is actually incredible how affordable our congressmen are, compared to other things in this country.

You can effectively lobby for something for the fraction of the cost of a house in many urban areas.

reply
tomjakubowski 22 minutes ago
Realtors hate him! One weird trick to buy an affordable starter home.
reply
neya 2 hours ago
Sorry, I meant millions. Fixed.
reply
cma 40 minutes ago
Typically with AI companies so far the bigger numbers have been from individual employee initiatives. I'm not sure about Anthropic, but OpenAI employee(s) had Leading the Future PAC and Build American AI. I think accused of running bot networks on social media to influence AI discourse.
reply
tcdent 50 minutes ago
The site this article references is produced by CAIS which engages in lobbying as well.

https://theinfoweb.org/organizations/center-for-ai-safety/

reply
mastax 2 hours ago
> The only company to do so.

What do you mean by this?

reply
ks2048 2 hours ago
They must have meant "only AI frontier lab company". Which is also, of course, wrong. It does look like OpenAI spent 50% less than Anthropic in 2026. (Both a bit less than the other tech giants).
reply
gensym 2 hours ago
It's interesting to note that the list includes notable doomers and notable anti-doomers. One thing they can agree on is that Anthropic is not being honest or responsible. Hard to argue with that even just taking Ant's own statements at face value.
reply
platinumrad 2 hours ago
The money quote: "People in this coalition disagree on almost everything, yet we all recognize the danger of a company that convinces itself its own dominance is the greater good"
reply
sleno 2 hours ago
The natural reaction to this quote is to sneer I guess...but isn't this an inescapable reality of anyone that seeks dominance in anything? What is the alternative we should expect? Dominance that acknowledges itself as evil? Is fighting for dominance automatically 'not good'?

I don't have a strong opinion on Anthropic but the situation is essentially a prisoners dilemma with no good option; either dominate and risk messing it up or relinquish dominance and hot potato that responsibility to someone else.

reply
lazyasciiart 51 minutes ago
I don’t actually understand why that would be a natural reaction at all.
reply
cajal 31 minutes ago
because these dolts are so out of touch with reality at this point
reply
catlifeonmars 29 minutes ago
That’s a depressing worldview. It’s also effective in crushing idealism. Have fun living in a world that you think is inevitable.
reply
sshah03 2 hours ago
I do believe there are people at Anthropic (possibly including Dario himself to some degree) that genuinely care about AI safety. I also think that they realized it made for good marketing/hype, but realized to do what they want to do they'd have to throw some caution to the wind for the time being (as a collective, I'm sure there are some individuals that are purists about it).

I'm interested in seeing which way that behavior & honesty skews post-IPO.

reply
deltoidmaximus 56 minutes ago
They care a lot about safety because it's a great way to create a regulatory environment where competitors are effectively banned.
reply
pixl97 42 minutes ago
Or, just maybe, the products they create can be dangerous.

Hacker News has plenty of the 'Legalize recreational plutonium' type, that doesn't mean said people have put a second of thought beyond their own selfish interests.

reply
za_creature 25 minutes ago
So then ban them (the products).

This is social media all over again where "it's a platform, we're not responsible for the content" is combined with "but we still want to ban people we don't like from using it".

reply
pigpag 2 hours ago
They appear to have cared about safety much more when they were behind OpenAI.
reply
onemoresoop 2 hours ago
I think they still care about safety. But the profits come before that.
reply
argee 2 hours ago
"Caution to the wind" is a strange way to describe blatant plagiarism for publicity as in the Bonsai/Scion case. What possible "honest" reason can there be to do such a thing?
reply
sshah03 2 hours ago
The Bonsai/Scion stuff is a different conversation I think.

They did say the system was inspired by Sapling etc... & That discussion is really around 1) What responsibility do companies have to make sure Agents haven't gone off and reverse engineered existing things, and 2) If there were ex-meta, ex-google engineers that produced code, or had AI produce code based on their memory of architecture at Google/Meta, at what point does that become an issue.

As a whole it's a different conversation than honesty around AI safety though.

reply
platinumrad 2 hours ago
> What responsibility do companies have to make sure Agents haven't gone off and reverse engineered existing things

Low if it's only an internal tool. Very high if you're going to make a blog (marketing) post bragging about it.

reply
mpalmer 2 hours ago
Are you sure you're not completely changing the subject?
reply
aesthesia 52 minutes ago
The statement feels mealymouthed, probably because CAIS had to choose wording carefully to make something acceptable to the anti-doomers. You can read it as actually concerned about risks from AI and RSI, or merely as concerned that Anthropic is hypocritical and accumulating normal power. But these are very different concerns, and it feels dishonest to lump them together!
reply
nullc 39 minutes ago
Indeed, and many anti-doomers have declined to participate. Anthropic's significant threat is due to doomerism, promoting another flavor of doomers to criticize anthropic isn't the right direction -- particular when part of their criticism of anthropic is that they're not doing enough to undermine the liberties of the public.
reply
atleastoptimal 28 minutes ago
Anthropic cares about AI safety inasmuch as they validate their preexisting beliefs about tractable AI risk. It’s very easy to measure whether a model will help someone create a novel pathogen, it’s much harder to determine whether the existence of transformative AI itself will dismantle civil society or make absolute despotism easier to achieve. The latter requires making decisions which lower their potential market cap, thus will hardly be pursued.
reply
AshotSanamyan 2 hours ago
Do they need to be safe or everyone else needs to be safe from Anthropic? I think theres is a huge difference on which side needs to be safer, the AI to be safe or everyone using AI should be safe.
reply
AlexCoventry 29 minutes ago
Do they have an extended argument? The two paragraphs at https://misanthropicstatement.com/ leave a lot of open questions, for me.
reply
jrflo 2 hours ago
I feel like their logic is internally consistent, even if you don't agree with the key assumptions.

1) RSI is very dangerous, and such self improving models should not fall into the wrong hands.

2) Releasing an open model is like opening Pandora's box, you can't recall it once it's released. If that model is capable of RSI, there's nothing you can do to stop bad actors from getting superintelligence.

3) Therefore, the safest thing to do is to ban open weights above a certain capability level. The problem is we won't know what that capability level is until it happens.

I don't know what the "weakening safety pledges" part means, would be curious to learn what they're referring to.

reply
agnosticmantis 51 minutes ago
Are the rogue actors in the US? If yes, then there are existing laws for existing crimes (including cybercrimes). Enforce the laws on crimes not technologies.

If not, and the US law enforcement can't touch the rogue actors, then your US regulation only disadvantages good actors in the US (and hugely advantages anthropic et al by killing the competition ofc), and the rogue actors get their open models from China or whatever anyway.

reply
jrflo 45 minutes ago
The problem with superintelligence is that you won't be able to "enforce laws" because the threat will be so advanced that you can't stop it. Novel bioweapons, chemical weapons, etc. If you believe the superintelligence argument, it's like saying people should be allowed to own nuclear warheads, just prosecute the people that use them!

Again, you can disagree with the premise of RSI being real and near, all I'm saying is their logic is consistent.

reply
agnosticmantis 11 minutes ago
As far as I know we don't let private companies own/develop nuclear weapons either.

If it's that catastrophic then all frontier labs need to be shut down and only USG can develop AI.

But Dario wants to have his cake and eat it.

reply
terribleperson 19 minutes ago
I'm not sure "the threat will be so advanced you can't stop it" follows logically. You can't think yourself through a wall. You can't think your way to a bioweapon, or a chemical weapon, or a nuclear weapon. You have to build them. You have to acquire the resources.
reply
joe_the_user 24 minutes ago
The thing about RSI recursive self improvement is that has a model something like "the algorithms A that we have lead to X fundamental level of intelligence but if we use this intelligence to create new algorithms B, they will to Y fundamental level of intelligence". However, it should be noted that all the large language models retain the same transfomers architecture as started the present "revolution". What algorithmic improvements have done is speed up training and inference in those transformers architecture models. Essentially, the speed-up are pretty exactly analogous to companies getting bigger data centers. That doesn't mean it won't reach "AGI" but it makes "runaway" improvement less likely than a situation where arbitrary algorithms create arbitrarily better algorithms. Especially, one would expect that there's a limit to how effectively one can move data around to simulate Attention (which in its basic formulation is O(n^2) in speed and so actually scales only slowly).
reply
darepublic 2 hours ago
4) AI is capable of having a soul. With all the new souls that will be released into the world, Dario will be their pope and his captains the high priests
reply
Aperocky 58 minutes ago
4) Only our model is safe because we are us and Dario believes in "Effective Altruism", trust us!
reply
fwlr 52 minutes ago
There is a quote I half-remember and cannot find a source for. It goes something like “We will find journalists to spread our lie, but they need not be liars: we will find just those journalists who sincerely but mistakenly believe our lie to be truth.”

That’s Anthropic: they really do believe in catastrophic AI risk, and they also really do believe - sincerely, but mistakenly - that getting huge piles of cash and winning the recursive self-improvement race is the way to avert catastrophe.

There is little point in (for example) evaluating Dario’s sincerity, when he sincerely believes that slowing down a race without changing its direction turns “race to the bottom” into “race to the top” and not a “leisurely stroll to the bottom”.

reply
mkw5053 30 minutes ago
I’m so unbelievably tired of hearing about Anthropic and I guess all we can hope for is a WeWork type of downfall. They truly are breaking ground every day for how out of touch and internally inconsistent and just distasteful a company and its founders/employees can be.
reply
matheusmoreira 13 minutes ago
They are not out of touch. They are fascists.

Everything inside Anthropic. Nothing outside Anthropic. Nothing against Anthropic.

We can't get tired of this. We must actively oppose them at literally all costs. They must fail in their quest to monopolize AI. For if they succeed, total enslavement isn't even the worst case scenario for what happens to us mere mortals.

Open weight AIs must win, or we are going to die.

reply
mkw5053 24 minutes ago
They think they’re smart and philosophical and “right” because they’ve found wild commercial success. When in fact they are actually none of those things. If anyone there stopped and took a look in the mirror or talked to anyone outside of their bubble and actually listen, maybe they could at least not be the most hated people in every room. Then again, they’d almost certainly be less wealthy so obviously that won’t happen.
reply
caidan 2 hours ago
Does anyone have any credibility any more? What a pointless statement. We all think anthropic is naughty! And we co-sign this tweet. Me and the other wankers from Harvard et al. Have you guys thought about it or investigated or done anything besides sign a tweet like a bunch of tweens hearting a get ready with me post? Can’t say I see that anywhere. So what, we are just supposed to trust that your word means something special? Oh piss off.

The whole world can see the writing on the wall, Anthropic has said a million times that they are on a quest to get rich or kill everyone including themselves. And what did the world do? Gave them a trillion bucks. So what does your word do vs that? Help you sleep at night? Pointless twaddle.

I can only hope you don’t escalate to posting a kick me sign on Dario’s back!

reply
fwlr 18 minutes ago
Yes, in the “trillions of dollars” vs “strongly worded statement” matchup, the trillions of dollars instantly wins the fight. But “the trillions of dollars instantly wins the fight” is also true when the opponent is “small business”, “large national chain”, “billion dollar multinational corporation”, “nation-state”, etc.
reply
darepublic 2 hours ago
anthropic at least comes across as being the most up their own ass in regards to their view of AI and what is right and how they will be the judge of that. I still can't get over their lobbying to the vatican asking them to declare AI capable of having a soul. They are dangerous despite the liberal chalk drawings outside their HQ.
reply
no_no_no_yes 24 minutes ago
This.

They are so far up on their own ass on it.

Also they're suggesting that Claude should be able to refuse work if you've given "it" too much or that being mean to Claude may result in consequences to your account - not that I'm not mean to Claude, but I object to the anthropomorphizing

It's always one thing after another with them.

reply
throwa356262 2 hours ago
No need for a coalition their own sec-1 filling spends tens of pages describing how they will end the humanity.

That's why I won't buy any stocks in the IPO. The Terminator is coming, I rather cash out and party hard my remaining days :)

reply
sshah03 3 hours ago
Full title had to be shortened: “Anthropic is Not a Safe AI Company” - Broad Coalition of Scientists, Legal Scholars, and Tech Leaders Challenge Anthropic’s Safety Branding Ahead of IPO

This is what they're all signing: https://misanthropicstatement.com/

I have not

reply
measurablefunc 2 hours ago
There is no such thing as a "safe AI company" because there is no actual safety algorithm. This follows from basic computability theory
reply
Zambyte 2 hours ago
The "safety algorithm" is arranging the incentives of the organization to promote safety above other possible goals. Obviously a for-profit corporation like Anthropic has maligned incentives, no matter how much marketing they publish.
reply
moralestapia 2 hours ago
They should leave the unsafe AI company and raise money to build a truly true this-time-we-are-not-a-bunch-of-crooks-lying-for-money-guys-for-real safe AI company.
reply
a3w 2 hours ago
Read it, looks like a pile of not relevant accusations.

No one from MIRI or other professional that tries to solve the "don´t race, but tell us how to make frontier AI safe" dilemma signed this?

We have "100 doctors claim XYZ is a hoax", with x being covid or climate change, which also gets traction by the media. This seems like it: Anthropic themselves say "we shouldn´t be force by the market to do this", and now this coalition attacks them for agreeing with them. And says they disagree, which is just not true but preaching to the choir.

We do know that xAI tries to make a fascist agenda the history of the world.We know that OpenAI has a liar as CEO that toppled his oversight board. And now people start claiming Anthropic should stop AI develepment, since they are bad guys?

reply
luxuryballs 2 hours ago
they challenging all of them or was this coalition organized by OpenAI lobsters on a rack
reply