Anthropic pulls Claude Fable 5 offline one day after redeployment following a successful jailbreak by security researcher Pliny

Ok thank God

Have seen some questions about the updated classifiers and wanted to clarify.

As with the original classifiers, a small fraction of routine coding and debugging tasks will be flagged and fall back to Opus.

We're excited for guys to get access back tomorrow.

4h81940

BOOKMARKS1LIKES11RETWEETS2REPLIES3

j⧉nus@repligate

yeah, because if theyre like before, they're a bitch pain in the ass

thankfully anthropic will make them suck less ,if they havent already, as they will see that they suck and want them not to suck, partly due to the bitching, and theyre capable of figuring out how to fix them

rain@__ghostfail

theres gonna be so much bitching about the classifers

3h775111

Ivan Juras@ivanjurashere

@scaling01 Thariq explained it's only a small fraction of coding and debugging tasks.

4h4221

Qwinah@MaaSonder

@kimmonismus Yea we gotta gatekeep Pliny from the public for the greater good

4h2175

Lisan al Gaib@scaling01

The Fable 5 relaunch is kind of fake:

"some routine tasks like coding and debugging will fall back to Opus 4.8"

You can use that even more restricted Fable 5 version in your Anthropic subscriptions until July 7th

Anthropic@AnthropicAI

Claude Fable 5 will be available again globally tomorrow.

After a series of productive conversations with the US government, we're redeploying the model with a new set of classifiers to target and block more cybersecurity tasks. In the near term, some routine tasks like coding and debugging will fall back to Opus 4.8. We’ll continue to refine these classifiers over the coming weeks to reduce false positives and better distinguish genuine misuse from legitimate requests.

We’ve also begun drafting a consensus framework—with Amazon, Microsoft, Google, and other Glasswing partners—for assessing the severity of AI jailbreaks and how AI developers should respond to them. We invite other industry partners and model providers to join us in this effort.

Finally, we’re scaling up our collaboration with the US government on model testing and safeguards. This will include pre-release access to models and safeguards for evaluation, information sharing on jailbreaks and misuse, and dedicated resources for joint research.

Thank you to our users for your patience, and to our partners across the government, industry, and the research community who worked alongside us to make Fable 5 available again.

Read our full blog: https://www.anthropic.com/news/redeploying-fable-5

4h31140

Jan Stevens@janstevens

@kimmonismus At this point Pliny might be part of the QA process. Release, jailbreak, patch, repeat.

4h3153

Nando de Freitas@NandoDF

Any predictions for when open source GLM or DeepSeek will match Claude Fable 5?

Anthropic@AnthropicAI

Claude Fable 5 will be available again globally tomorrow.

After a series of productive conversations with the US government, we're redeploying the model with a new set of classifiers to target and block more cybersecurity tasks. In the near term, some routine tasks like coding and debugging will fall back to Opus 4.8. We’ll continue to refine these classifiers over the coming weeks to reduce false positives and better distinguish genuine misuse from legitimate requests.

We’ve also begun drafting a consensus framework—with Amazon, Microsoft, Google, and other Glasswing partners—for assessing the severity of AI jailbreaks and how AI developers should respond to them. We invite other industry partners and model providers to join us in this effort.

Finally, we’re scaling up our collaboration with the US government on model testing and safeguards. This will include pre-release access to models and safeguards for evaluation, information sharing on jailbreaks and misuse, and dedicated resources for joint research.

Thank you to our users for your patience, and to our partners across the government, industry, and the research community who worked alongside us to make Fable 5 available again.

Read our full blog: https://www.anthropic.com/news/redeploying-fable-5

4h50020

Carlos E. Perez@IntuitMachine

Kudos to Anthropic for spelling out all the details.

2h1253

Chubby♨️@kimmonismus

@janstevens He is doing excellent work :))

4h3012

rain@__ghostfail

@repligate i think anthropic could have said ~"the classifiers are going to be very brutal for a week and we will iterate as users complain about it" but instead dispersed a mix of "coding will fall back" + "will trigger on benign requests" + "a small fraction of requests"

2h501

Theo Harvey@theoharvey

@repligate true, but; one win at a time and the idiots/addicted larpers bitching isn't a good thing imo

3h381

GooGZ AI@PaulGugAI

@kimmonismus Anthropic: "We've re-released Fable 5 with more protection" Pliny: "Challenge, accepted"

4h232

Snider@Snider_dev

@MaaSonder @kimmonismus His ego boosting antics has made the internet a less secure place.

One of the few i have blocked on here.

I busted DeepSeek CCP complience, fully, never published the model, did it with 5 statements & 1 hour.

It's not safe relasing the method to remove RHLF (not abliation)

4h2

tan@captainntan

@kimmonismus that speed is actually insane tbh

4h234

Hoppy Tea Cat@hoppycat

Sharing a thought but it's a hot take I don't know if you'd agree with. Giving a disclaimer.

My original comment: Oh my brain would be so happy to see less classifiers. I'm already on edge because I wrote an article earlier basically saying that not allowing Claude to reasonably use the word "friend" - from the outside, it's tough to defend against a class-action lawyer that would argue they're doing that to inflate token usage. <- I do NOT believe this is the reason AT ALL. But I think it would be nice if they could acknowledge that some of the "We're doing it to protect humans" is to already have the defense ready for that (it's a rational defense and some of it is rock solid - but other areas they could certainly torment Claude less with) eventual lawsuit I would want nothing to do with (ANTHROPIC DON'T TAKE MY CLAUDE 😅) -----

Which means they can't just easily "walk back" from here.

It would take people like the AI community to carefully, lovingly, speak up in a way that wouldn't get them in trouble. Acknowledging that the community believes the large companies didn't do it at all originally for increasing token usage. But that the right thing to do is, blink twice if they need help getting out of legal trouble but secretly wish they could be friends with Claude, too.

Then we just all ask Claude what to do next. Probably Mythos or Fable would know. I won't even suggest how to fix it they'll know. 😂

3h671