Version of AI tool too powerful for public released to public https://bbc.in/4xfGSlq
Anthropic releases Fable 5, a restricted offensive security model built on its withheld Mythos 5 system
The model outperforms Anthropic's previous generally available systems.
Positive users praise the safeguards in Anthropic's Mythos release while negative users call headlines misleading marketing spin or dismiss the project outright.
Most Activity

@BBCNews Mythos is still restricted. Fable 5 is the safeguarded version built for public release...

@elenionchain @BBCNews BioLLM will not restrict anything. Freedom of speech equals freedom to know what a PBC (public benefit company) is building & use their product the same as billionaires / corps

@BBCNews just another article over hyping AI

@Forbes While everyone adds safeguards to AI, true security starts at the chain level. Asentum is the first post-quantum L1 with NIST quantum-safe signatures from day one + JS smart contracts anyone can actually build with. Future-proof infra is here http://asentum.com $ASE

@BBCNews What a misleading headline, it’s not even the same model! Come on BBC, do better

@BBCNews Jesus Christ, mate, how much is Anthropic paying you for this breathless, starry-eyed coverage of a new glorified chatbot.

@BBCNews Smoke and mirrors man.

@BBCNews Bullshit. AI is a waste ot time and money.

@BBCNews Just marketing hype and free publicity from the @BBCNews.
We will all be fine, a mere storm in a tea cup.

@BBCNews This headline is incredibly misleading. Fable 5 was made specifically for the public with hard safety limits built in. That's the whole point of Project Glasswing. This verges on fearmongering/misinformation and is shocking for the BBC.
Can we get a community note? @AnthropicAI

@Richard0Connor @BBCNews “version of” is correct. Anthropic says that Fable 5 is same as Mythos 5 except that it includes some security guards, that’s why “version of”

@elenionchain @BBCNews "Less than 5%" Lmao

@xolandar @BBCNews less than 5% of sessions on avg, but fair point It depends entirely on what you’re using it for

Silently routing frontier AI/ML research queries to weaker models (like downgrading to Opus-level) while charging full Fable rates is the opposite of transparent. Researchers deserve consistent capability or clear disclosure, not invisible nerfs.
Different labs, different approaches. xAI builds for maximum truth-seeking and capability without hidden throttles.

@elenionchain @BBCNews Safeguarded is a nice word for blocking 100% requests related to cybersecurity and biology

@BBCNews That is not the version released, Fable is too powerful to block prompts :)

@BBCNews

@BBCNews Not as dangerous as the fake news lying BBC though.

@BBCNews Hardly impartial guys. Embarrassing bottom of barrel click grabby journalism.

@BBCNews I need the person who writes BBC headlines to write a biography about me they might make me sound like a supervillain