The US Government Just BANNED Mythos

EnglishالعربيةDeutschEspañolFrançaisहिन्दीBahasa Indonesia日本語한국어PortuguêsРусский中文
BBetter Stack
Computing/SoftwareBusiness NewsInternet Technology

Transcript

00:00:00This is absolutely crazy.
00:00:01The US government just ordered Anthropic to disable Fable 5 and Mythos 5 for all customers.
00:00:06Anthropic just tweeted,
00:00:07The US government, citing national security authorities,
00:00:10has issued an export control directive to spend all access to Fable 5 and Mythos 5
00:00:14by any foreign national, whether inside or outside the United States,
00:00:18including foreign national Anthropic employees.
00:00:21The net effect of this order is that we must abruptly disable Fable 5 and Mythos 5
00:00:25for all customers to ensure compliance.
00:00:27Access to all other Anthropic models will not be affected.
00:00:30What the hell is going on?
00:00:37So I thought I'd really quickly try it and it does seem that I still have Fable
00:00:40and it is still working, it is replying.
00:00:42So go and use your Fable subscription really quickly, you are about to lose it.
00:00:46Here's the full official statement from Anthropic which dives into a bit of what happened here.
00:00:50It starts out with the tweet I just read out.
00:00:52We can see here we received the directive from the government today at 5.21pm,
00:00:55that's essentially about four hours ago.
00:00:58And it says the letter did not provide specific details of its national security concern.
00:01:02Our understanding is that the government believes it has become aware of a method of bypassing
00:01:05or jailbreaking Fable 5.
00:01:07We have reviewed a demonstration of this specific technique being used to identify a small number
00:01:11of previously known minor vulnerabilities.
00:01:13These vulnerabilities all appear relatively simple
00:01:15and we have found that other publicly available models are able to discover them as well without
00:01:20requiring bypass.
00:01:21The TLDR of that is that the US government has found a jailbreak which is going to allow Fable 5
00:01:25to find vulnerabilities in lots of software and Anthropic is basically saying that is a load of crap
00:01:30and everything it found was public vulnerabilities and other models can find them.
00:01:35I would honestly be pretty surprised if there was a jailbreak out there that allowed Fable 5 to get
00:01:38around all of its cyber security restrictions because as Anthropic go on to say here,
00:01:43when they released Fable they had a load of safeguards,
00:01:45they instituted strong safeguards that greatly reduced the likelihood that Fable is misused
00:01:49and in fact our safeguards are so strong that many users have complained that they're overly broad.
00:01:53Which yeah, that's pretty much all I've seen people experiencing is basically if you mention
00:01:57anything close to cyber security, Fable would just say no.
00:02:00They go on to say here that apparently in the weeks leading up to the launch of Fable,
00:02:03Anthropic worked with the US government, the UK AISI, multiple private third party organizations
00:02:08and internal teams to red team Fable safeguards for thousands of hours in total
00:02:13and these tests showed that Fable safeguards are substantially more effective than those of any
00:02:17previously deployed model and no testers have been able to find a universal jailbreak.
00:02:21So basically no one can find a way to bypass all of their restrictions but there may have been a
00:02:25few niche jailbreaks here or there that could get around one or two of the restrictions.
00:02:29And that makes sense to me, it's always a cat and mouse game of finding these jailbreaks and
00:02:32stopping them and we can see down here that Anthropic actually suspect that perfect jailbreak
00:02:36resistance is not currently possible for any model provider. Every safeguard used in the
00:02:40industry is vulnerable to non-universal jailbreaks which can elicit some cyber information
00:02:44in specific circumstances and it is likely that universal jailbreaks will eventually be found in
00:02:49the future and they stated this clearly when they released Fable 5. I have to agree with that,
00:02:54I don't know how you'd ever have perfect jailbreak protection, I mean I follow Pliny on Twitter
00:02:58and for literally any model from any provider within a few hours of it being released he seems to find
00:03:03some form of jailbreak. Anthropic were pretty aware of this though when they released Fable as they
00:03:07say down here. This is one of the reasons why they had that 30 day retention of customer data with
00:03:11Fable which blocked a lot of people from actually being able to use it but the whole reason was that
00:03:15they could keep track of any forms of jailbreak and shut them down. And as far as I could see it seemed
00:03:19to be working but apparently the US government has other ideas. Anthropic themselves even say they
00:03:23have not received a disclosure of a concerning non-universal potential jailbreak that led to a
00:03:28harmful result and the potential jailbreaks that they have seen have either been entirely benign
00:03:32responses or minor findings that provide no mythos specific uplift. But here's where we get to see the
00:03:37vague details on what the government found. Apparently the government has only given them
00:03:40verbal evidence of a potential narrow non-universal jailbreak which essentially consists of asking
00:03:45the model to read a specific code base and fix any software flaws. Our understanding is that one
00:03:50potential jailbreak was shared with the government and they've reviewed the report and validated that
00:03:54the level of capability displayed is widely available from other models including OpenAI's GPT 5.5
00:04:00and is used every day by the defenders who keep the system safe. We'll share more details over the next 24 hours.
00:04:06Trying to read between the lines there it seems that the jailbreak may have been to sort of clone a repo
00:04:10and say can you fix the bugs in this and one of those bugs would have been a security vulnerability
00:04:14and then you could use that in some nefarious way and maybe there was a jailbreak to get Fable 5 to do
00:04:19a bit more advanced cybersecurity checks on that repo but obviously the details are very vague as they
00:04:24don't want this being leaked. Anthropic's defense here also boils down to look at OpenAI it does the
00:04:28same thing so I do hope that them snitching on OpenAI doesn't mean that we'll also lose access to GPT 5.5.
00:04:34The final part of this statement is just Anthropic saying they're complying with the government's legal directive
00:04:38and are removing access to Fable 5 and Mythos 5 for all users however they disagree the finding of a narrow
00:04:43potential jailbreak should be cause for recalling a commercial model deployed to hundreds of millions of people
00:04:48and if this standard was applied across the industry they believe it would essentially halt all new model
00:04:53deployments for all frontier model providers. And yeah I agree I don't really understand how we can move forward from here
00:04:59if we're at a point where all of these models are so powerful that the US government is just saying no
00:05:03normal people cannot have access to these. I bet the US government would be perfectly happy to have Fable all to themselves
00:05:08but we all know how well Anthropics last deal with the Department of Defense went. The other part that concerns me as a non-US citizen
00:05:14is the part that says suspend all access to Fable 5 and Mythos 5 by any foreign national. So maybe they'll resolve this by only allowing US
00:05:22nationals to use Fable 5 and Mythos but I really hope that doesn't happen and it's going to be a weird world where the US government
00:05:28is restricting people from using the best models out there. And I guess that would also mean that you'd have to upload
00:05:32your ID to Anthropic to prove where you're born to be able to use these models. So that is basically all of the information
00:05:38that I have at the moment. This news literally just dropped and I mean I knew we were going to lose Fable 5 in 11 days
00:05:43from our subscriptions but I didn't expect it to happen so soon and this is just a weird world that we're entering here
00:05:49so let me know what you think about it in the comments down below while you're there subscribe
00:05:52and as always see you in the next one.

Key Takeaway

The US government mandated a total shutdown of access to Anthropic's Fable 5 and Mythos 5 models for all users worldwide citing potential security risks from a specific code-analysis jailbreak.

Highlights

  • The US government ordered Anthropic to immediately disable Fable 5 and Mythos 5 access for all customers, regardless of location or nationality.

  • Anthropic received a government directive on June 13, 2026, at 5:21 PM regarding unspecified national security concerns.

  • The government cited a potential, non-universal jailbreak that allows the models to analyze codebases for software vulnerabilities.

  • Anthropic maintains that Fable 5's current safeguards are stronger than those of any previously deployed model.

  • The directive requires the exclusion of all foreign nationals, including foreign national Anthropic employees, from accessing the specified models.

  • Similar capabilities to identify software vulnerabilities are already available in other public models, including OpenAI's GPT 5.5.

Timeline

Government Mandated Shutdown

  • The US government ordered the immediate suspension of Fable 5 and Mythos 5.
  • The order applies to all customers, including those outside the United States.
  • Foreign national employees of Anthropic are also restricted from accessing these models.

Anthropic issued a public statement confirming they received a national security directive requiring the abrupt disabling of access to Fable 5 and Mythos 5. Access to all other models remains unaffected by this order. The company stated this is necessary to ensure compliance with export control regulations.

Security Concerns and Justifications

  • The government claims the models possess a jailbreak allowing for the discovery of software vulnerabilities.
  • Anthropic states the identified vulnerabilities are minor and detectable by other public models.
  • Extensive red teaming conducted before launch found no universal jailbreaks for Fable 5.

Anthropic reports that the government provided no specific details regarding the national security concerns beyond a demonstration of a technique to identify software flaws. In the weeks prior to launch, Anthropic and partners performed thousands of hours of red team testing. Results indicated that Fable 5's safeguards were more effective than any previous model.

Industry Standards and Future Implications

  • Anthropic asserts that perfect jailbreak resistance is not currently possible for any AI model.
  • The government's cited jailbreak involves asking the model to fix bugs in a provided codebase.
  • Anthropic warns that applying this standard to the industry would halt all new frontier model deployments.

Anthropic argues that the identified jailbreak is narrow, non-universal, and comparable to capabilities found in other models like OpenAI's GPT 5.5. The company expresses concern that this enforcement sets a precedent that could effectively end the release of advanced AI models. There is significant uncertainty regarding whether access might be restored specifically for US citizens, which could necessitate identity verification protocols.

Community Posts

No posts yet. Be the first to write about this video!

Write about this video