Overactive safety filters on Fable 5 1M systematically pause benign chats and downgrade users to Opus 4.8

had an idea for a big fable project, set it up, and let it cook

came back an hour later and it had triggered the safeguards and fell back to 4.8 10 minutes in

back to codex 😬

I love being too dangerous for an AI to even talk to

Good news everyone, kernel methods are AGI, pack it up

It's getting ridiculously Anthropic. Nothing even remotely problematic was asked.

@AnthropicAI Omg, I told it that I'm not a microbiologist and it should forget all of that and my Fable access came back

LMFAO this also caused it to kill all my other Fable chats. Absolute fucking amateur hour from @AnthropicAI

the first time i ran this it got to michael levin's bioelectricity work and blocked itself

@baym try @AskVenice

@jon_barron 😂😂😂

@baym We chose to make a tradeoff and are working on refining classifiers and broadening access.

I had no idea I was working on something so dangerous 🥸

His name was William Tyndale

@danshipper DM'd for more info! we're working on tuning the balance here more post-launch :)

@danshipper Me too, thats why I spend my whole day to figure out how to bypass it, just using fable to beat Fable : https://github.com/HughYau/anti-fable

@liqsweep Fable 5 asked to ask them about mantis shrimps

Opus 4 asked

Fable 5 started writing about mantis shrimps, and the classifier started firing

How does this even work

@baym

@baym omg this is the tweet that makes me try Fable - and YES indeed I am in the cool squad

@kimmonismus Its only beginning, think anthropic making it clear their god-level AI is not for peasants like us to freely. By end of the year we will need to send application and have interview before we are even allowed to their frontier models at all.

is it the flowers???

@baym Uggg........working on this. I'm so sorry for this mess of a roll out and messaging for biology.

But for now...the mocking if fair.