Claude Fable 5: Anthropic Warns the House Is on Fire, Then Sells Tickets to the Bonfire
“Five days after begging the industry to hit the brakes, Anthropic shipped the most powerful public model in history: safety as a press release, capability as a product.”
Sally's not done with you yet.
Drop a URL, screenshot, or file and Sally will give you the honest truth.
Sally's Take
Let us savor the timing, because it is exquisite. On June 4, Anthropic published a paper titled When AI builds itself, a genuinely alarmed plea for the entire industry to agree on a coordinated brake pedal before models start improving themselves without us. Recursive self improvement, they warned. Existential pacing problem, they said. Five days later, on June 9, they released Claude Fable 5, the first publicly available Mythos class model and, by their own benchmarks, the most powerful AI any member of the public has ever been allowed to touch. The fire alarm and the gasoline arrived in the same week, shipped by the same people, signed by the same brand. You cannot script this. You can only stand back and admire the nerve.
To be clear, the product is real and it is frightening in the good way. Fable 5 posts state of the art results on nearly every benchmark that matters: software engineering, vision, knowledge work. Stripe ran it on a fifty million line codebase and compressed a migration that would have taken a team over two months into a single day. That is not a marketing gradient, that is a structural shift in what one engineer can do before lunch. Pricing landed at 10 dollars per million input tokens and 50 per million output, less than half what Mythos Preview cost, which means the frontier got dramatically cheaper the same week it got dramatically more capable. Half the AI tools on the internet, this cynical little roast engine included, quietly got smarter overnight without sending you a memo.
Then there is the two tier world they built, which is the most honest thing in the whole launch. Fable 5 is the same model as Mythos 5 with the safeguards bolted on: three classifier systems for cybersecurity, biology and chemistry, and distillation, all of which fall back to Claude Opus 4.8 when you wander somewhere dangerous. That fallback triggers in under five percent of sessions, which Anthropic frames as reassurance and which a cynic reads as an admission that ninety five percent of the time you are talking to the unfiltered thing. Meanwhile the actual full power Mythos 5, the one with the guardrails removed, goes only to approved cyberdefenders and infrastructure providers. So the world now officially has a safe version for the public and a real version for the chosen few, and Anthropic told you that to your face while warning you the technology is too dangerous. The candor is the scariest part.
Think your work can survive this?
Drop a URL, screenshot, or file and Sally will give you the honest truth.
What Actually Happened
- •Anthropic launched Claude Fable 5 on June 9, 2026, its first publicly available Mythos class model, generally available and state of the art on nearly every benchmark.
- •Mythos 5 is the same underlying model with key safeguards removed, restricted to approved organizations only: cyberdefenders, infrastructure providers, and select researchers.
- •Fable 5 carries three classifier safeguards for cybersecurity, biology and chemistry, and distillation, falling back to Claude Opus 4.8 in under five percent of sessions.
- •Pricing is 10 dollars per million input tokens and 50 per million output, less than half the cost of Claude Mythos Preview. Free on Pro, Max, Team and Enterprise plans until June 22.
- •It shipped five days after Anthropic published a paper warning that frontier AI is advancing too fast and urging the industry to coordinate a slowdown.
Who Got Burned
Anthropic's own messaging took the first hit, because it is hard to sell a global slowdown on Monday and the most powerful public model in history on the following Tuesday without somebody noticing the calendar. Every rival lab that ever got lectured about safety culture now has a permanent screenshot. Engineering teams that spent two months budgeting a migration got burned by a tool that does it in a day, which is wonderful until you remember a tool that fast does not need a team that large. And anyone clinging to the comfort that the truly dangerous models stay locked away got a polite reminder that the locked away version and the public version are the same brain, separated only by a classifier that steps aside ninety five percent of the time.
Silver Lining
Strip away the irony and what remains is genuinely the best argument for the technology and against panicking about it in the same breath. Anthropic could have shipped Fable 5 with a triumphant capabilities video and zero context. Instead they published the danger paper first, built the safeguards in public, documented the fallback rate honestly, and kept full Mythos behind approval gates for defenders rather than selling it to the highest bidder. That is not a company that forgot its own warning. That is a company betting that the only way to make powerful AI safe is to ship it carefully and out loud rather than let someone less careful ship it quietly. The price halved, the capability doubled, and for once the safety story is not a footnote, it is the headline. If the frontier has to move, this is roughly what moving responsibly looks like, contradictions and all.
Your turn. Drop something.
Drop a URL, screenshot, or file and Sally will give you the honest truth.