Trump admin loves to do crazy shit on Friday nights. In this case, it's issuing "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 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.
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.
When they tried to do this to encryption, the cypherpunks managed to pull off a win. I wonder if it will go the same way this time round.
maga maoism
Once they release a smarter model, it's so lame working with the last one. They are kind of dull.
I kept Opus as the default (and was playing with the thought of hardcoding 4.7 selection to get rid of some of the 4.8 regressions).
Fable cost is prohibitive, and it has more false positives for me than Opus 4.7. It feels a bit like the same outcome from a year ago where 4.x was having a ton of regressions vs 3.7 (though this may not be as steep.) Maybe a 5.5 in 4-6 months will bring the real improvement, like 4.5 did versus 4.1.
It communicated in better detail than Opus. That's what I miss.
Do you have the replies saved? Opus on xhigh or up can be instructed more easily to adjust output (and less default template triggering)
I assume they're in the logs. I'm still not at your levels of wizardry yet. I'm still using off-the-shelf harness/leashes/skills for the most part.
You can often tell it to do things you want in your prompt. Even just as a sentence it works 99.99% of the time (except on
<=higheffort, then it will just suck.)I got blessed by limits probably, stopped at 15% of my weekly limit and I've been using it a lot.
Sad that it went away, it could do much more in 5 minutes than opus.
Are you using it interactively?
I had it map the stacker news' codebase with n agents in parallel, some telegram bots and all-day interactively with its vscode extension.
The curious part was that the vscode extension wasn't participating in the limits at all, they would only go up with claude code cli.
Must be I call
claude -pand they penalize me for that? Or usingxhigh/max? Not sure. Either way, moot point now, lol. I'm hapy I didn't tune my framework to it.Jailbreakers were Amazon researchers.
source: https://www.washingtonexaminer.com/policy/technology/4607706/amazon-researcher-anthropic-ai-security-weaknesses/
That
mightexplains why I just lost access to Fable in the middle of a chat.It's funny because I was telling someone yesterday that Fable is sensitive to emotive conjugation. It blocked me when I asked it to "find bugs," but assisted when I asked it to "harden the code."
I might be growing a paranoid, but my first thought was that OpenAI is behind this.
It's hard to flag this as paranoia when it's that obvious that OpenAI is behind this.
Well, the obvious thing is that Anthropic didn't play ball with the US government earlier this year (#1443175, #1444092, #1444092, #1438014).
However, I feel like the Anthropic guy was constantly running around saying their models were so dangerous and so powerful that they were going to end the world. So, while I'm hopeful that there will be a wave of public sentiment against such bans as this, I won't be surprised if it doesn't materialize.
Seems like we may be on the way to some kind of nationalizing of the big AI companies.
And now I can't pull up the announcement anymore:
if it is OpenAI, Anthropic looks like they are tattling on them:
I'm a big fan of Anthropic printing and publishing the source code to train Mythos in book form. They can also just upload it to HF. Problem solved. World "saved".
Seriously though; they asked for this, through their endless wolf-crying weak sauce. Good that they admit they have no moat, and GPT can do the same.
The timing here is interesting. We're seeing AI become load-bearing infrastructure in security contexts — smart contract auditing, code review, vulnerability detection — right as the political environment starts treating powerful models as controlled exports. If you're running AI-assisted security tooling, this has direct implications: access to the frontier models that make automated vulnerability detection actually useful may become geographically restricted. Already thinking about fallback options.
David Sacks on the subject
https://x.com/i/status/2065853007619588171
They wanted regulation, W