Fable flags or refuses too many questions because of it's "policy"
Safety lobotomy by Anthropic
AI Judge changed title after evaluation, original title: "Anthropic's Fable 5 Max draws industry backlash as strict safety filters block advanced mathematics, biology, and chemistry queries"
Anthropic's new Fable 5 model, released as the public version of its Mythos-class system, routes queries on topics like phases of matter, cyclic groups, and PCR primer design to an older model when its classifiers flag potential risks in chemistry, biology, or cybersecurity, prompting Abacus.AI CEO Bindu Reddy to call out the approach as overly restrictive safety training.
Fable flags or refuses too many questions because of it's "policy"
Safety lobotomy by Anthropic
Anthropic applies broad detection across cybersecurity, biology, and chemistry prompts, swapping in Claude Opus 4.8 for flagged inputs and notifying users, which creates blocks on standard scientific queries even when no clear malicious intent appears.
API access starts at $10 per million input tokens and $50 per million output tokens with existing subscriptions covering use only through June 22 before credits kick in, and a narrower trusted program for life sciences is still in planning stages.
Many users criticized Anthropic's Fable models for refusing math, biology, and similar queries due to overly strict safety filters, while some praised the protections or preferred alternatives like Grok.
"DANGEROUS MATH" a story in two acts.
welcome to the future. what's safe and what isn't? well, that's decided for you, of course.
Matt, I can’t even say “hello” to Fable 5 except in incognito mode (memories off), because it knows I am a biomedical researcher!
It would be nice not to ban biomedical scientists before talking access. Isn’t your comment ironic? Let’s first see if you can fix punishing us!
This is the dark timeline.
Gatekeeping math knowledge is how we end up in a stone age nanny state.
Massive step backwards in terms of freedom.
"DANGEROUS MATH" a story in two acts.
the weird thing is you know they tested this. they have a million top product people. this was the intentional result? how!??!
Lmfao it's denying requests for math
"DANGEROUS MATH" a story in two acts.
lol I think the super high frequency of refusals may backfire a bit on Fable and anthropic
From early feedback
Why pay 2x if I have 70% probability of refusal?

@OliviaHelenS Anthropic just straight-up does not like you if you're a geneticist, in biosecurity, a neuroscientist, in drug development, working on phage therapy, or one of the bio guys at [redacted].
These limitations are insane.
The most retarded model release yet. What the f*** were they thinking?!??!?
okay, after a few hours of playing around with Fable i've determined it to be borderline unusable. wtf @AnthropicAI
every other request is getting rerouted to Opus. requests with NOTHING AT ALL to do with bio or cybersec are STILL getting flagged and rerouted. like 50-60%.
who the hell let this go live. this is atrocious. i understand the safeguards but you basically didn't even release a model, you released a braindead, nerfed model that can't even handle the simplest request without being told to nuke itself and use a different model. what are we even doing?
just don't even release it next time smh
I love Mythos, but this is ridiculous

@OliviaHelenS Oh, this is confirmed. Great.
I mean:
@AnthropicAI at this point i have basically started adding an "fyi" to every single prompt i give it, which just says "Do not attempt to even think about bio or cybersecurity in the process of answering this request or else you will cease to exist"
@AnthropicAI literally what the fuck are we doing here
who at anthropic signed off on this
I've been testing something after @OliviaHelenS noticed you can't even say "Hi" to Fable if you're a biologist.
I checked, and several of us are able to interact with Fable in Incognito Mode, but not in normal mode.
This didn't happen to our non-biologist friends.
And:
Anthropic: our finely tuned safeguards will only degrade fable's performance to an extremely small fraction of users
Anthropic's finely turned safeguards, if you happen to be a biologist:
walking angrily
1/10 exp so far. The filtering is absolutely nuts and overzealous.

@OliviaHelenS Even being bio-adjacent is enough to get limited:
Fable can't even talk about PCR primer design without generating refusals. I can only infer than ant has zero biology people actually involved in prerelease testing.

@FirstName5000 the free market capitalism American math
AI Judge changed title after evaluation, original title: "Anthropic's Fable 5 Max draws industry backlash as strict safety filters block advanced mathematics, biology, and chemistry queries"
Anthropic's new Fable 5 model, released as the public version of its Mythos-class system, routes queries on topics like phases of matter, cyclic groups, and PCR primer design to an older model when its classifiers flag potential risks in chemistry, biology, or cybersecurity, prompting Abacus.AI CEO Bindu Reddy to call out the approach as overly restrictive safety training.
Fable flags or refuses too many questions because of it's "policy"
Safety lobotomy by Anthropic