Science Government forces Anthropic to pull its latest AI model because it's too good.

https://archive.ph/SNAjZ


Supposedly the US government has forced Anthropic to freeze their latest AI model because it's so good that it can hack the US government.


Statement on the US government directive to suspend access to Fable 5 and Mythos 5​

Jun 12, 2026
The US government, citing national security authorities, has issued an export control directive to suspend all access to Fable 5 and Mythos 5 by any foreign national, whether inside or outside the United States, including foreign national Anthropic employees. The net effect of this order is that we must abruptly disable Fable 5 and Mythos 5 for all our customers to ensure compliance. Access to all other Anthropic models will not be affected.

We received the directive from the government today at 5:21pm (ET). The letter did not provide specific details of its national security concern. Our understanding is that the government believes it has become aware of a method of bypassing, or “jailbreaking” Fable 5. We reviewed a demonstration of this specific technique being used to identify a small number of previously known, minor vulnerabilities. These vulnerabilities all appear relatively simple, and we have found that other publicly-available models are able to discover them as well without requiring a bypass.

Anthropic’s posture with respect to Fable’s safeguards, as laid out in our launch blog post, is the following:

  • We have instituted strong safeguards that greatly reduce the likelihood that Fable is misused for tasks related to cybersecurity (among others). In fact, our safeguards are so strong that many users have complained that they are overly broad.
  • In the weeks leading up to the launch of Fable, Anthropic worked with the US government, the UK AISI, multiple private third-party organizations and internal teams to red-team Fable’s safeguards for thousands of hours in total.
  • These tests showed that Fable’s safeguards are substantially more effective than those of any previously deployed model.
  • No testers have yet been able to find a universal jailbreak—a jailbreak method that can very broadly bypass the model’s safeguards, unblocking a wide range of cyber capabilities.
  • We suspect that perfect jailbreak resistance is not currently possible for any model provider. Every safeguard used in the industry is vulnerable to non-universal jailbreaks (which can elicit some cyber information in specific circumstances), and it is likely that universal jailbreaks will eventually be found in the future. We stated this clearly when we released Fable 5.
  • Given that perfect jailbreak resistance does not appear to be possible today, Anthropic adopted a defense in depth strategy with Fable 5. We aimed to make jailbreaks either narrow (in the case of non-universal jailbreaks) or very expensive to produce (in the case of universal jailbreaks), and to combine this with thorough monitoring to quickly detect and shut down any successful attacks. This is also why Anthropic has required 30-day retention of customer data with Fable—a policy change that carries real costs for us with customers, but that allows us to research and mitigate jailbreaks.
  • We stand by this defense in depth strategy. It reduces the risks posed by Fable, making them comparable to the risks of existing models already deployed across the industry.
  • We have not even received a disclosure of a concerning non-universal potential jailbreak that led to a harmful result. The potential jailbreaks that have been disclosed to us are either entirely benign responses or are minor findings that provide no Mythos-specific uplift.
To date, the government has only given us verbal evidence of a potential narrow, non-universal jailbreak, which essentially consists of asking the model to read a specific codebase and fix any software flaws. Our understanding is that one potential jailbreak was shared with the government. We have reviewed a report that we believe is the basis of the government's directive and validated that the level of capability displayed there is widely available from other models (including OpenAI’s GPT-5.5), and is used every day by the defenders who keep systems safe. We will share more details over the next 24 hours.

We are complying with the government’s legal directive and are removing access to Fable 5 and Mythos 5 for all users. However, we disagree that the finding of a narrow potential jailbreak should be cause for recalling a commercial model deployed to hundreds of millions of people. If this standard was applied across the industry, we believe it would essentially halt all new model deployments for all frontier model providers.

As we have stated publicly, we believe the government should have the ability to block unsafe deployments, as part of a statutory process that is transparent, fair, clear, and grounded in technical facts. This action does not adhere to those principles.

We apologize for this disruption to our customers. We believe this is a misunderstanding and are working to restore access as soon as possible.
 
Última edición por un moderador:
Or maybe it just proves that the people the government hires are incompetent. It's like when they referred to Edward Snowden as a 'computer wizard' without actually qualifying what they meant or providing examples of his talent.
 
Or maybe it just proves that the people the government hires are incompetent. It's like when they referred to Edward Snowden as a 'computer wizard' without actually qualifying what they meant or providing examples of his talent.
They are right. He was a computer warlock
 
So, apparently what happened is, some Amazon employee tried some jailbreaking techniques with Fable, and after repeated attempts, got *something* to work. (Finding some code vulnerability or some such shit.) Either that employee, or that employee's supervisor, decided to inform the US government about what they found, instead of reporting it to Anthropic. (Keep in mind Amazon is Anthropic's biggest investor.) That then prompted the U.S. Department of Commerce (specifically the Bureau of Industry and Security) to issue an "Export Control Directive" (via a FAX of all things, and was supposedly sent by Trump-appointed secretary of commerce Howard Lutnick himself) ordering Anthropic **NOT** to "take down" Fable/Mythos, but rather insure that anyone not born in the USA can not access Fable/Mythos. Many of Anthropic's employees and consultants are not born in the USA. There is also the web and labyrinth of web VPN's and proxies (including residential proxy services) to consider. There's no way for them to filter this immediately, so they just took it down. If it comes back, then they will probably require users to upload their driver's license or other form of ID to them, or some shit like that, which will piss off privacy advocates (which is like most people, at least I would like to believe). It's unclear how Anthropic will have to deal with their coders and employees who are not USA born... does this mean they can no longer work on anything for Anthropic?

For context, an "Export Control Directive" is usually used for things like uranium refinement equipment, weapons control systems... things like that. Which means, an AI model is now considered by the US government to be in this class. The implications of that are pretty profound.

One thing is probably true, is that whoever at Amazon reported this to the government instead of reporting it to Anthropic, is probably out of a job, and might be in pretty big trouble as well, as billionaires don't like their shit getting fucked with, especially in such an epic way.
 
Última edición:
Or maybe it just proves that the people the government hires are incompetent. It's like when they referred to Edward Snowden as a 'computer wizard' without actually qualifying what they meant or providing examples of his talent.
This should be obvious by now. The average government employee is an utter moron who knows nothing about technology. Its why its a blessing Elon, Sacks, etc have gotten involved and brought in people with an actual understanding of technology.
example: https://ndstudio.gov/posts/automating-federal-retirements
2 startup engineers were able to complete a project in 2 months the government had been failing for 7 years because they were too retarded to build a basic web app.

In regards to the main topic though, this is probably a good thing if done right and will likely be a key step in guaranteeing American Hegemony. People are seething because Karpathy isn't a US citizen currently, but they can just fasttrack the handful of serious AI researchers to citizenship to make sure they keep building in the US. Anthropic has already spent the past year bitching about the Chinese distilling their models, this is just a step towards actually taking action on it. If the US is able to establish a monopoly on top level AI labs, we will realistically be able to take all of the top level talent (which is a very small pool) and have them work here. It can catapult the US so far ahead of the rest of the world they won't be able to compete.

So, apparently what happened is, some Amazon employee tried some jailbreaking techniques with Fable, and after repeated attempts, got *something* to work. (Finding some code vulnerability or some such shit.) Either that employee, or that employee's supervisor, decided to inform the US government about what they found, instead of reporting it to Anthropic. (Keep in mind Amazon is Anthropic's biggest investor.) That then prompted the U.S. Department of Commerce (specifically the Bureau of Industry and Security) to issue an "Export Control Directive" (via a FAX of all things) ordering Anthropic **NOT** to "take down" Fable/Mythos, but rather insure that anyone not born in the USA can not access Fable/Mythos. Many of Anthropic's employees and consultants are not born in the USA. There is also the web and labyrinth of web VPN's and proxies (including residential proxy services) to consider. There's no way for them to filter this immediately, so they just took it down. If it comes back, then they will probably require users to upload their driver's license or other form of ID to them, or some shit like that, which will piss off privacy advocates (which is like most people, at least I would like to believe). It's unclear how Anthropic will have to deal with their coders and employees who are not USA born... does this mean they can no longer work on anything for Anthropic?

For context, an "Export Control Directive" is usually used for things like uranium refinement equipment, weapons control systems... things like that. Which means, an AI model is now considered by the US government to be in this class. The implications of that are pretty profound.

One thing is probably true, is that whoever at Amazon reported this to the government instead of reporting it to Anthropic, is probably out of a job, and might be in pretty big trouble as well, as billionaires don't like their shit getting fucked with, especially in such an epic way.
1781378962389.png
It sounds to me like they reported it to both Anthropic and the government. I wouldn't be surprised if there were existing deals in the background in regards to these models as well. Anthropic's issues with the DoW likely led to some scrutiny for partners like Amazon who have heavy government contracts. Alot of government Claude usage is through AWS bedrock, and govcloud probably has stipulations in place that require Amazon to report any vulnerabilities in their services directly to the feds.
 
Última edición:
This reminds me of the days in the 90s when PPC was breathing its last dying gasp and Apple made stupid ads about how their computers were so powerful that the federal government treated them like weapons! ... ignoring the fact that they just put them on the same list of "Don't sell this to North Korea, Cuba, Iraq, or Iran" computer parts as they always had.

Furthermore is it just me or is there a bit of childishness about this? Like Altman at a school lunch table saying "UH, WELL, UM, I MADE A UH COMPUTER AI SO...SO SMART IT THE COMPUTER THEY HOOKED IT TO FOUND EVERYONE'S PASSWORDS? AND, UH, THE GOVERNMENT MADE THEM UNHOOK IT? BECAUSE IT WAS SO SMART?"
 
The below was posted to Reddit (https://archive.ph/wip/FJhxQ); not sure if it's an AI response or not:


UPDATE — Fable 5 / Mythos 5 shutdown. Amazon was reportedly the trigger, and the story keeps getting weirder. Consolidated, sourced rundown for anyone who lost access.

WHAT WE KNOW ABOUT THE TRIGGER

It was reportedly Amazon. Per the WSJ (corroborated by Axios, TechCrunch, Reuters, The Information), Amazon researchers jailbroke Fable 5, and CEO Andy Jassy took the findings directly to Treasury Secretary Scott Bessent and other officials. National cyber director Sean Cairncross convened White House staff Friday morning, and Trump signed off. At least five other companies reportedly raised concerns too. Anthropic says it had notified the government of the June 9 release multiple times with no prior objection.

WHY THIS IS STRANGE

Amazon isn't a random third party. It's one of Anthropic's largest investors, holds a board seat, provides AWS + Trainium chips, and distributes Claude via Bedrock — with Anthropic reportedly committed to ~$100B in cloud spend. So the company's own biggest backer flagged its flagship model to the government.

No official motive given. Two plausible readings, not mutually exclusive:
  • Defensive/contractual: AWS serves government clients and was responding to an admin request for feedback. Flagging a vuln in a model AWS itself sells is arguably liability coverage, not sabotage.
  • Competitive: Amazon builds its own models (Nova) and has an interest in no single supplier — even one it's invested in — becoming indispensable or a risk liability.
Worth noting: per the WSJ, the jailbreak was "a series of prompts" to get the model to surface info usable in cyberattacks. Katie Moussouris (Luta Security), who reviewed the report Anthropic shared, called the government's reaction disproportionate to what's actually in it.

THE PUBLIC FIGHT

David Sacks (former admin "AI czar," now PCAST co-chair) accused Anthropic of being "reckless," claiming the government asked Amodei to fix the jailbreak or pull the model and he refused. A source close to Amodei denied that to the NYT, saying Anthropic was open to discussing it. Two conflicting accounts, and the government's technical report has never been made public.

STATUS RIGHT NOW
  • Fable 5 and Mythos 5 still disabled everywhere, including Bedrock. No restoration timeline.
  • Opus 4.8 and all other models unaffected.
  • Prediction markets (Polymarket, Kalshi) price ~68–71% odds of Fable 5 access restored by July 1.
  • Pre-IPO valuation took a small hit (Hyperliquid perp down ~3.7%, from >$1,800 to ~$1,627). Anthropic filed its confidential S-1 on June 1; the shutdown is the kind of thing that ends up in a prospectus risk-factors section.
If you were building on Fable 5, Opus 4.8 is the practical fallback for most SWE work. Curious whether anyone here actually hit the classifier fallback to Opus during those three days, and how noticeable the capability gap was.

Sources: WSJ, Axios, TechCrunch, Reuters, The Information, NYT, Anthropic's own statement.
 

The Scoop​

The White House imposed export controls on Anthropic’s powerful Mythos AI model partly over suspicions that a China-linked group had accessed it, a person familiar with the matter said.

The Trump administration on Friday directed Anthropic to limit access to Mythos, and its consumer version, known as Fable 5, to US citizens. That meant that even Anthropic employees who are foreign nationals would have been prohibited from using the models. Anthropic chose to remove the models from the market completely.

Anthropic tightly controlled the release of Mythos, which was launched in April, limiting access to a select group of companies that could use it to plug security holes. The plan was to ensure its capabilities wouldn’t make it into the hands of hackers. Anthropic has said Mythos represents a danger to the public because of its ability to find bugs in computer code, which could be exploited by malicious actors.

It’s unclear how the White House learned of the issue, which organization accessed the model, and how it gained access to Mythos.
But if the Chinese government had access to Mythos, it could pose national security risks to the US. China could also attempt to reverse-engineer and copy the model in a process known as distillation.

On Saturday, an adviser to President Trump, David Sacks, in a post on X shed new light on the events leading up to the White House decision to implement the export controls.

Sacks said that the government received a warning that Fable 5 could be jailbroken. When the Trump Administration notified Anthropic of the jailbreak, Anthropic co-founder and CEO Dario Amodei said the jailbreak was not a serious risk and refused to fix it, Sacks alleged.

Anthropic released Fable 5 to the public earlier this week — a version of Mythos with guardrails that prevent its use for cybersecurity purposes. A jailbreak is a way to remove the guardrails of an AI model.

A person close to the White House said Amazon informed the government about the jailbreak, and that its CEO, Andy Jassy, had been in contact with members of the administration about it.

Sacks said the administration reluctantly issued the export controls, which required Anthropic to only allow US citizens access to Mythos and Fable. Those requirements caused Anthropic to revoke access to the models for everyone.

An Anthropic spokesperson said the White House didn’t raise Chinese access to Mythos in its conversations around the Fable jailbreak and export controls. The AI company prohibits access to its products from within China.

An Amazon spokesperson said “it’s not uncommon for governments to seek our counsel on potential security risks. When they occur, we don’t share the details of these discussions.”

The White House and Sacks did not immediately respond to requests for comment. The Information first reported that Jassy had contacted the White House.

Know More​

“In the past, Anthropic has always said that safety must be top priority and taken super seriously. In this case, Anthropic prioritized the continued offering of the consumer model over safety,” Sacks wrote in his tweet.

It isn’t the first time Anthropic has clashed with the White House. It lobbied against the administration’s efforts to preempt state AI regulation, and it is suing the Pentagon after reaching an impasse over the use of its models for autonomous weapons.

Sacks says those previous conflicts did not factor into the administration’s decision on Mythos. “Those trying to misdirect and tie this action to the prior DoW/Anthropic issues are wrong. The Admin values Anthropic’s technical capabilities and feels that this issue, while serious, should be easily resolved. The ball is in Anthropic’s court,” he said in the X post.
 
Atrás
Top Abajo