The US government just killed billions in Anthropic revenue over a non-universal jailbreak that exposes capabilities that other models like GPT-5.5 also possess
US government orders Anthropic to suspend foreign access to Fable 5 and Mythos 5 over a code-analysis jailbreak
Anthropic claims these capabilities are already available in GPT-5.5.
Many users criticized the US government blocking Anthropic revenue over the jailbreak as an unacceptable slowdown of American AI progress that benefits China while also faulting Anthropic for courting regulation and overhype.
Most Activity
This was all allegedly triggered by a Mythos jailbreak that was shared with the US Government. This is Anthropic's response:
'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.'
ANTHROPIC NEW ROLE ALERT: Anthropic is currently searching for an American typist for @_sholtodouglas, so he can still access Mythos to continue researching.
actually might be billions in revenue if this goes on for a few weeks
The US government just killed billions in Anthropic revenue over a non-universal jailbreak that exposes capabilities that other models like GPT-5.5 also possess

Anthropic is disabling Fable 5 for everyone. All users. It's going to be a fight.
Apparently Mythos 5 specifically is being placed under special restriction by the Pentagon and will not be allowed for use even by most Government employees. Restricted to agencies only.
The US government just killed billions in Anthropic revenue over a non-universal jailbreak that exposes capabilities that other models like GPT-5.5 also possess
Just occurred to me that Anthropic employees who are not US persons will not be able to use Fable/Mythos, making this plausibly (and to be clear, accidentally) the first regulation on recursive self-improvement.
> 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)
You asked to be regulated by people who don’t know the difference.
You fucked around and found out.
This was all allegedly triggered by a Mythos jailbreak that was shared with the US Government. This is Anthropic's response:
'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.'
*billions in revenue if they sustain this restriction for like a few weeks+
The US government just killed billions in Anthropic revenue over a non-universal jailbreak that exposes capabilities that other models like GPT-5.5 also possess
@scaling01 more drama, but much stupider?
this might be better than the Sam Altman / Board drama in late 2023 or the supply chain risk designation saga
https://www.anthropic.com/news/fable-mythos-access
The US government just killed billions in Anthropic revenue over a non-universal jailbreak that exposes capabilities that other models like GPT-5.5 also possess

@scaling01 The OAI lobbying effort sends their regards
last ARR was $47B or $3.9B/month or $130M/day this is still there as it's all from old models
Fable would have probably been something like ~$2B/month
actually might be billions in revenue if this goes on for a few weeks

@AndrewCurran_ Yes, other models may be able to find *those specific vulnerabilities*. But Dario, you’ve just spent a month shouting at how many more deep exploits your model can find. So when someone says they can jailbreak it, all bets are off, regardless of the examples *they* give

@OwnerEmeritus @AndrewCurran_ your scope is off. what you are suggesting is they disclose that this is capability is not possible or probable . if gpt 5.5 pro does it , then even if your pro govt the US govt has to block gpt 5.5 pro .. and it has to halt Us progress and investment on AI .. which would cause

@scaling01 this slows down US progress in ai, completely unacceptable

@OwnerEmeritus @AndrewCurran_ ok so your ok with the government killing a commercial enterprise and favoring the slower alternative ? which is one release or 2 away ? 3-6mo.. open might not be as close.. scope . its not just what anthropic says at that point in time. its not just what anthropic says

@scaling01 well, Dario dramatically overhyped it, so it’s on him that his marketing and fundraising strategy is now biting his butt

I get what you’re saying. But Anthropic’s premise here is that the vulnerabilities found that forced the move “really aren’t that bad and other models found them anyway”. Which in a vacuum aligns what what you’re saying and I agree.
But it’s not in a vacuum. Anthropic has been yelling that Mythos can find 100x the vulnerabilities as humans, and figure out how to string together multiple minor vulnerabilities to create huge breaks. So the shutdown isn’t so much coming from what vulnerabilities the jailbreaks found. It’s coming because the model that the creator says is capable of soooo much has now been demonstrably jailbroken, which would mean sky’s the limit for the jailbreakers.

@AndrewCurran_ Le sigh.