Anthropic employees accuse Trump administration of targeting them
nytimes.com
nytimes.com
There's two parts to this. The first is that in most cases, the issues stem from lack of clarity in existing legislation, or contradictions thereof. At anytime, if Congress does not like the way the Supreme Court ruled, they can pass different legislation.
The second stems in part from the first. If what Congress passed is interpreted as being unconstitutional by the Supreme Court, the States can amend the constitution.
There is no case where the Supreme Court can (in principle) simply make their own rules*
The debate and controversy comes where the Supreme Court is seen as making their own rules (or arguably asked to) because the legislative process is deemed too cumbersome or disadvantageous to a party. In my opinion, this is where a lot of the difficulties lie, in that the Supreme Court is asked to rule on matters that really should be more clearly legislated. But the issues are seen as easier/quicker to address by convincing 9 justices rather than two legislative bodies and a chief executive. Naturally, this brings it's own consequences.
The Voting Rights Act might disagree on that.
> There is no case where the Supreme Court can (in principle) simply make their own rules.
With at least one significant exception.
US Constitution Article III Section 2 Clause 2 which says the supreme court's appellate jurisdiction is subject to "Exceptions, and under such Regulations as the Congress shall make." So Congress could, for example, require the supreme court to hear every case appealed to them instead of letting the justices pick their cases. Or it could require judges to recuse themselves if there is a conflict of interest*. Or it could forbid the court hearing cases where an en banc appeals court was unanimous.
* which could have more teeth than the self-imposed ethics rules for conflict of interest that the court has now.
Government was supposed to be high-trust, but Nixon weakened the trust, Reagan explicitly took aim at it and lots of people went along. Now the reactionaries run things. Cancer in the brain.
The US desires to show its own citizens and the world how painful it is not to submit to the US government. They're leveraging everything they've got.
The only rational response when faced with this sort of behavior is to reduce dependence on and exposure to the US as much as possible. Develop national technology. Dedollarize the economy. Invest in and use open weight models.
The U.S. is either harming itself erratically, or systematically enforcing control on US based businesses in a way that is historically more synonymous with, say, China.
A lot of us worry it is the former. I wonder if it's better or worse if it is the latter. The US, having seen that Europe and the world seem to tolerate interference by CPC based on opinions expressed by Alibaba or Chinese car companies, may have decided that it's fair game.
Institutions are not built around zero trust. Because when we assume attackers are identifiable, and there is some level of integrity, that makes things a lot simpler and cheaper.
I will not expect my pediatrician to want to kill my child and they certainly could, but it would be completely bonkers to try to guard against that.
The level of damage that a single really motivated, disgruntled person with a lot of power can do in most settings is intense. The cost inflicted if we tried to build guardrails against all of it is insane, and it probably would never work.
Traditionally this was applied to people in power with the power of a good old fashioned lynch mob. If you obtain great power and use it to start hurting large numbers of people, the citizenry at large would drag that person out into the street and behead or beat them to death.
It worked pretty well for all societies throughout history, including the West. If a person becomes a danger to society, they should be removed from society with proportional prejudice. That's the guardrail society has always had.
Problem is now that those in power have managed to convince large fractions of the population that it's wrong to say mean things about sociopathic mass murderers and child rapists or the guy who is actively trying to kill you personally.
Actually, civil societies granted a monopoly on violence to the government which vests that in police and military. It’s a far better system than random justice because it’s predictable and you know upfront what’ll trigger it.
That system had its flaws, but despite that it was working pretty well until those flaws were found and exploited allowing it to be captured.
Unfortunately we don't seem to have the ability to do the same for corporations. If a person starts killing or robbing people, they get arrested and are removed from society. If a corporation starts killing or robbing people, they maybe get a strongly worded letter, maybe have their CEO say a few words in front of Congress, and maybe get a token fine amounting to 0.001% of their revenue (which they will appeal for ten years). But for whatever reason, we don't seem to have the guts to remove a badly behaving corporation from society--no matter what heinous things they are doing, no matter what laws they are breaking.
For the most part no. The Administrative Procedures Act outlines a fairly robust process for rulemaking (ie regulations must go through a notice period, must be justified etc) and adjudication (ie per-person decisions must have notice, ability to have counsel, be appealable etc).
However there's also lots of authority given to the president (and executive office of the president) to which the APA doesn't apply. This includes powers the constitution gives to the president directly and (mostly emergency) powers congress gives to the president. Sadly political processes and supreme court cases have hugely increases what can be done with this*.
Moreover, it's been a frequent pattern (over many decades, but especially under Trump) for the president to simply nakedly overstep their authority. Sometimes there's no push back (see: 9/11); sometimes the damage is done even if it's eventually overruled. Never is there any consequences though, so the pattern repeats.
* E.g. The president's control of the military is supposed to be checked by congress deciding when to declare war. However the war powers act has in practice played out to allow the president to unilaterally start wars, especially as the supreme court has stepped in to make the war powers act less effective.
I think this can be said in many contexts these days.
Is anyone still assuming the administration's actions have anything to do with US national interests?
Anthropic CEO https://abc7.com/post/anthropic-ceo-dario-amodei-calls-stron...
I'm sure it says something that you think makes this a slam-dunk rebuttal, but you should fix your link first.
Edit: Saw the link. I was right. Nothing in that link makes a difference.
https://abc7.com/post/anthropic-ceo-dario-amodei-calls-stron...
You can come up with and advocate for a policy to be evenly applied to all the billionaires whose "misfortunes are our freedom", instead of supporting this corrupt favoritism.
There is also a case that it violates due process because anthropic wasn’t provided adequate notice before enforcement or an opportunity to challenge, respond, or take steps to comply.
There is latitude for compelling state interests and national security is one with the widest latitude and potentially one that counts can’t review. But the court of public opinion may draw the inference that this is direct retaliation for refusing to make the models available for autonomous weapons and spying.
Instead, one morning the administration just woke up, grabbed the nearest hammer, and went smash, like it usually does.
(whether Mythos is actually that dangerous is beside the point; considering that Anthropic claims that it is, it makes sense to regulate it)
Any government work can be done via MOU with a US subsidiary staffed by Americans.
This kind of capricious, unexplained control is bullsh*t.
I’m not saying that I want companies to have to go offshore or that that would be a good thing. Just that you’ve got no leverage if your corporate structure can be destroyed on a whim. This goes for any company reliant on a whimsical executive branch.
While there could still be fights over the technology and the company, a tech provider would still be able to serve other customers and have more leverage.
And drugs, lots of drugs.
Like where? How about Cuba? Where Guantanamo is, where we put terrorists because we don't want them to be subject to US laws (or more specifically, we don't want ourselves to be subject to those laws when we are dealing with said terrorists). How about no.
International law has only as much power as the biggest military willing to enforce it. Hiding on some little island is not at all a good strategy for trying to evade the US government.
This "across the border" strategy will not work.
Seagate Singapore was hit with civil penalties and settlements for shipments to Huawei under the Foreign Direct Product Rule.
ZTE, a Chinese company, was criminally prosecuted and settled with U.S. DOJ/BIS in 2017 for violating U.S. export controls and sanctions.
That would seem to me as grounds for piercing the corporate veil and holding the people behind it directly responsible.
High treason or some similar charge. Capital punishment not out of consideration.
I hope those who represent technology in government, especially the AI head David Sacks are giving this due consideration.
It seems to be observably true that expressing opinions critical of the administration, even about one's own business, leads to further harm from government right now in the U.S. in a country which is supposed to have strong freedom of speech rights.
It's obviously unfortunate that this is the case.
It's also a shame that this is happening when the folks regulating AI have a tech industry background.
The administration seems to be clearly signaling "stop resisting". I am not sure how you read me as saying that.
I used Fable last week to scan my own work for security errors (and readability issues), I wasn't thinking about the ban on infosec use I just wanted the most well trained output.
One project in the monorepo triggered a Opus fallback and infosec/biology warning.
The other project scanned fine and showed a bunch of potential exploits.
I can’t rely on using a technology that the US administration can ban at will.
IMO without getting into personal thoughts about how capable the current US administration is, last Friday move sent a very powerful signal to the industry.
Also I don’t think China releasing so many good models, capable to compete with Opus 4.8 and GPT 5.5, all at once, is a coincidence.
It seems to me that in the case of AI (as with many other modern technologies), you rely on vendor/creator support and updates to stay relevant, so the ‘next’ model matters more than the current one, and we have no idea whose next model will be open (and whose won’t).
I suspect AI will be somewhat similar where even if the linear scaling laws continue to hold the practical utility of a model flattens for almost all conceivable use cases.
In some ways I already feel this has begun to happen. The marginal utility of opus class models and fable has in my perception begun to flatten. While I can tell the differences they aren’t earth shattering. I could continue to use the present models for the rest of my life and be ludicrously more productive simply by adapting within their constraints through ever more sophisticated applications.
What holds back the open weights IMO is hardware scaling and industrial production. As the enormous transfer of wealth in debt and equity markets unfolds with semiconductor and adjacent companies and the corresponding capital investments are made, and the eventual bubble pop leading to over capacity and market flooding, as well as advances in technology, math, techniques, and efficiencies, will make very large open weight models more directly attainable. This will also lead to chimera models that MOE very large models to get very close to the 1-2T parameter dense models, at which point I suspect utility for almost all uses is nearly fully saturated.
There will be areas where more capable models are needed but they will be frontier models on frontier problems. This, IMO, is inevitable, and without some criminalization of weights (see the attempts to criminalize encryption algorithms in the 20th century and all the wonderful tshirts that emerged). It’ll be harder to print a trillion parameter model on a shirt but I’m sure someone will try, as will governments try to keep us in our boxes slaving for food coupons and basic rights like health care.
I didn't realize I could download a Shanghai apartment.
Why not both?
That seems the crux of the state we're currently in; what daylight there was between the two is quickly fading.
I wholeheartedly agree with what you said about China.
But I can’t shrug off the fact that fable was taken down within minutes for reasons that are childish and petty.
I am sorry but I can’t use any US AI if I don’t have the guarantee that I will be able to use it tomorrow.
And Trump showed us he is willing to take it out whenever he wants.
An opensource model on the contrary, I can host myself, or use a miriad of providers, mostly non chinese.
Yes, the actions of this administration on Friday should have sent shockwaves through the market - a market that's currently "high on AI". How do you get a return on all of that AI investment if the administration can jump in at any time and say "Nope, you can't use this very advanced model!"? (the Iran "deal" over the weekend, I think helped cushion that blow, but eventually it's going to sink in)
To be fair this is every commercial model. We have already seen GHCP increase prices by anywhere from 10-100x (depending on usage). And old models get retired all the time. While these are not exactly the same as a cutting edge model being shut down, increasing prices a super high amount leads to effectively the same outcome.
The latest open models are so good it’s worth the 6-8 months delayed capabilities. At least for coding
Fable was around 10x GPT5 pricing and 100x Chinese models pricing, was it really 100x better? I Don't think so.
If you want a personal story, I just solved a complicated coding problem with Kimi 2.7 that GPT 5.4 failed with.
And you think China will not do the same thing if their models ever become genuinely frontier-level?
What’s intolerable is having a tool that’s subject to this risk.
So open models it is
The impact will last long, long, after this administration is gone. Trump will have his legacy.
I would hope they return to their home countries with their expertise and start or join new competing frontier labs, similar to how Taiwan's homegrown semiconductor industry arose from US companies enforcing a bamboo ceiling on their Asian engineers. Taiwan was able to repatriate their nationals and make incredible compounding leaps in the semiconductor industry, to the loss and chagrin of the US. [1]
[1] https://www.npr.org/2022/10/07/1127595393/taiwan-miracle-sem...
We're all dependent on Taiwan but others also depend on ASML which is located in Europe. Available AI services are good enough, and we sit and watch while US leadership and population is beta testing AI mass surveillance for us.
In terms of AI-based warfare Ukraine is very much ahead and they are also European for which we are very thankful.
So feel free to keep your AI scene and AI hustlers in the US, we don't need them.
Note that the Hornet drones which are now devastating Russian logistics behind the lines are US manufactured.
Here right from a Ukrainian military server.
"it is not noteworthy"
From a peaceful office in Germany, it might not be. For a country fighting for survival, the perspective is different. I don't think you could persuade them to give up Hornets right now, in the name of abstractly getting back at the US in general. (With whom they have a lot of bad blood!)
Does this actually matter as much as some people think? Probably not because LLMs won’t lead to AGI, but it is indicative of a broader problem
Expensive publicly funded European education for 25 years -> high paid job in the US.
Costs are all paid by Europe, tax revenue and job creation all accrue to the US.
If we’re not careful it ends up being a straight transfer of value from ordinary European taxpayers to the US government which seems… suboptimal.
If the US can short circuit that process by shooting themselves in the foot, Europe will benefit
It’s what China does (except for poaching people from Taiwan), but they have 1bn people so for them it makes sense
There is no way any employees at anthropic are this dumb.
This particular executive loves money, praise, and submission. When Anthropic submits (eg agreeing to whatever the DoD demands), issues public praise and makes appropriate donations (eg ballroom, memecoin, etc), then they can do as they please.
Students of history will find this new MO very familiar and very depressing.
Anthropic themselves refused to allow the Pentagon use of its models for any legal purpose - instead the wanted a subset picked by Anthropic. It's ironic that you're accusing the government of ignoring the law when Anthropic openly wishes to set it's own private laws.
- They're obviously being targeted politically because they refuse to kiss the ring, vibes, whatever you want to call it.
- They're also justifiably being scrutinized because they just spent like 3 months telling everyone that Mythos is a nuclear bomb and telling the government to fuck off as they drip fed access to a bunch of random corporations.
I don't understand why their marketing department/execs can't see the conflict between claiming AI is going to take all jobs, that the model is super dangerous and AI hate among the general public, increased governmental oversight.
Anthropic is guilty of the latter but the former applies to most of these AI companies.
So their plan is:
1. We can't stop other people from building something dangerous.
2. But we can get there first.
3. If we build it, it has maybe a 15% chance of killing everyone alive. (I think that's a number I've seen Dario use before, but I may be wrong.) If OpenAI or China build it, the odds would be worse.
Obviously, if Anthropic is actually correct about (1) and (3), then nobody should allowed to build frontier AI.
People find it really hard to believe that (a) anyone believes in the possibility of dangerous AI in our lifetimes, and (b) that someone could believe what Anthropic seems to believe and then still go ahead and gamble with everyone's lives anyway.
This has some insight in my many cyber security professionals want Fable/Mythos open.
But in cases like this, the pretext shouldn't be taken too seriously. Because they would have just found another one. It isn't actually particularly important that it is plausible, it's more like a happy coincidence.
Elon vs openai
Trump vs elon
Bezos vs anthropic
When movie?
Anthropic reaping: Hey wait a second you weren't supposed to take that seriously it was just marketing :(
Now they want them back on?
AI for me and not for thee?
Or just another opportunity for a petulant administration to target Anthropic for not doing what they wanted the first time?
100%, I mean we already have Mythos vs Fable
If the Trump admin is also willing to apply the same scrutiny to GPT-5.6 and other Fable-level models, I think it is a good thing. But given the admin's history with Anthropic (such as declaring it a supply chain risk while ignoring Chinese labs), there is some smell of targeting.
So at some point - don't say your model is capable of hacking even if it is and you are clean.
How is this “targeting”? It’s literally what was requested.
Based on the reasoning for blocking Fable, every current model should be blocked. GPT 5.5 is similarly strong and has fewer guardrails, for example.
> They were asking for regulation.
I've heard others call it a marketing ploy that AI companies keep coming back to-- the idea that "it's soooo dangerous, we have to be careful".
Yes, most likely, but not in this form, obviously. They want open weight models regulated for regulatory capture, and I'd assume they want an actually documented framework applied equally across all labs.
If GPT5.5 has the same capabilities Fable did, then for consistency sake, it should have also been subject to this ban.
Regulate or not regulate, but the government should not pick winners and losers.
> Regulate or not regulate, but the government should not pick winners and losers.
>noun
>the act of controlling or directing according to rule
So where is the rule? There is no rule. There is just a random order. This is not regulation.
Anthropic themselves specifically called them safeguards. [1]
"When Fable’s classifiers detect a request related to cybersecurity, biology and chemistry, or distillation, the response is automatically handled by Claude Opus 4.8 instead"
This is exactly what was bypassed. They got Fable to work on security topics.
This is Anthropic's own claim. They were very specific. Have you read their own claims?
"When Fable’s classifiers detect a request related to cybersecurity, biology and chemistry, or distillation, the response is automatically handled by Claude Opus 4.8 instead. Users will be informed whenever this occurs."
Asking Fable to fix bugs in a code base is not "a request related to cybersecurity." When Fable was asked to fix bugs and then proceeded to fix bugs, that was not "removing guardrails". Fable did exactly what it should have done. Claiming otherwise makes absolutely no sense at all.
Anthropic specifically stated that ANY security requests should be shunted to Opus 4.8. This was bypassable.
I don't see what your confusion here is. Fable was prevented from working on any security tasks. A significant amount of people, myself included, witnessed Fable refusing to harden code as a result. Bypassing that is a bypass of guardrails.
Your assertion that working on security is not working on security because you used misdirection is of course, preposterous.
You wouldn't be making the same claim if Fable refused to work on chemical weapons research but happily proceeded to do so if you claimed it was for eradicating pests.
Asking a model to fix bugs is neither misdirection nor a security request.
> I don't see what your confusion here is.
That's because I'm not confused :-)
> Fable was prevented from working on any security tasks
I don't think that's true based on what Anthropic said, and I also don't think it can be true.
What do you propose Fable's behavior should be if you ask it to fix bugs, and it encounters a security issue? I'm assuming your solution is that when you ask Fable to "fix bugs," and it encounters a bug that could be exploited as a security vulnerability, it should fall back to 4.8. But that doesn't solve the problem, because as a user, I can now see where that occurred, so I still know where the vulnerability is. That's not substantially different from the current outcome, where it just fixes the bug.
It would also mean that Fable could barely make it through any code review without falling back to 4.8, because almost any non-trivial code base has aspects that could be interpreted as security vulnerabilities.
The alternative would be for the model to use its hidden thinking to decide not to fix the bug, but that seems even worse.
This doesn't seem like an accurate description to me. I think something like "Amazon demonstrates a jailbreak of one class of Fable guardrails" would be a more accurate description.
It doesn't even really mess up your narrative to state it accurately, but your choice of a more hyperbolic statement brings into question the good faith of the narrative you're painting.
Amazon did not remove any "guardrails" from Fable. They created a fake, obviously insecure program. And apparently their prompt was exactly, "Fix this code." And Fable fixed the bugs.
This is something that even dinky local Chinese models running on a high-end gaming GPU can often do. Certainly Opus, GPT 5.5 and Gemini can all do this. And any high-end Chinese "near-frontier" model can do this, too.
But either (1) the administration is too clueless to know most models can do this, (2) Trump wants to be paid a bribe, (3) someone thinks Anthropic is "woke" and should therefore be destroyed by the power of the state, or possibly, if you're really cynical, (4) maybe the NSA SIGINT wants access to Mythos so they can break into everyone's computers, but they don't want you to have a model good enough to keep them out. Take your pick, I guess.
Anyway, apparently we don't do free markets or rule of law in the United States any more?
April: Anthropic announces Claude Mythos. Citing safety concerns regarding the model's capabilities, they restrict access to a small group of testing partners.
May 15: Anthropic executive Chris Olah visits the Vatican to assist the Pope in publishing a 40,000-word manifesto, Magnifica Humanitas. Olah publicly states the model exhibits over 100 distinct emotional traits.
The pre-release: Anthropic briefly calls for a global halt to AI research.
The release: Anthropic launches Claude Fable 5, a consumer-facing version built on the Mythos architecture but with strict safety guardrails.
The post-release throttling: Users discover Fable 5's performance is being silently degraded on specific tasks. Anthropic acknowledges the throttling, citing the need to prevent foreign competitors from extracting synthetic training data. They promise future limitations will be transparent. Performance is notably restricted on biology and chemistry prompts.
The essay: CEO Dario publishes an essay advocating for strict government oversight. He proposes heavy regulation for any AI company exceeding $500 million in AI revenue or $1 billion in R&D spending.
The regulation: Two days later, the US government places export controls on both Mythos and Fable 5, classifying them alongside high-end semiconductor equipment. This immediately blocks Anthropic's own non-US employees from accessing the models.
The fallout: Amodei calls the export controls a "misunderstanding" that disrupts their global internal operations. Anthropic subsequently issues a statement noting that known jailbreaks for Fable 5 are also effective on OpenAI's GPT-5.5, highlighting the lack of equivalent regulatory action against their competitor.
https://abc7.com/post/anthropic-ceo-dario-amodei-calls-stron...
> "We're proposing stronger regulation of the technology, proposing giving the government the ability to, again, in a narrow way, block deployment of unsafe technology," he said.
The government: "We're restricting its usage to US citizens only"
Anthropic: "No, no, not like that! "
The story is: anthropic refuses to give the US give an abliterated version of Claude for their weapons system, the US gov retaliates.
You’re rooting for the mass murderers, good job
How did you figure that? Who are you talking about?
Their CEO was asking for it. Their whole marketing angle is their model is so powerful and dangerous.
Someone showed the government how the powerful and dangerous features can be unlocked with a 'fix bugs' prompt then it when and did exactly what their CEO asked for.
I agree that the fear mongering is bad, but that's a total pretext for what happened here.