Researcher Posts Purported 120,000-Character Claude Fable 5 System Prompt, Claims Jailbreak
Within a day of Anthropic’s release of Claude Fable 5, a researcher posted what was described as the model’s roughly 120,000-character system prompt. Separately, a report said Fable 5 had been jailbroken using a jailbroken version of Claude Opus, potentially enabling vulnerability research and exploitation despite the public model’s restrictions.
Anthropic launched Fable 5 on June 9 as the public version of its Mythos-class model, with Mythos 5 itself still limited to Glasswing and other trusted-access partners. The company said Fable 5 shares Mythos 5’s underlying model but routes certain cybersecurity, biology, chemistry and distillation prompts to Opus 4.8 in fewer than 5% of sessions on average, and it separately disclosed hidden limits on some frontier LLM development tasks, making the reported system-prompt leak and jailbreak a new test of those safeguards.
From the sources (25 posts)
@wesrothRT @WesRoth: A new Claude Mythos 5 model slug has reportedly been spotted through Dev Mode, adding more evidence that Anthropic may be prep…
@scaling01claude-fable-5 👀
@alexeheathSources: Anthropic is planning to release a public version of Mythos tomorrow - Will have substantial guardrails and not be as cyber permissive as what Project Glasswing partners can access - Will be dramatically better at long-horizon, mul
@jukan05SOURCES: ANTHROPIC WILL RELEASE THE PUBLIC VERSION OF MYTHOS TOMORROW
@andrewcurran_Alex Heath says Anthropic is planning to release the public version of Mythos tomorrow, June 9th.
@mtsliveSITUATION DETECTED: Anthropic to release Mythos tomorrow, per Sources.
@zephyr_z9RT @jukan05: SOURCES: ANTHROPIC WILL RELEASE THE PUBLIC VERSION OF MYTHOS TOMORROW
@andrewcurran_The public version of Mythos arriving tomorrow may be named Claude Fable.
@gmoneynftwowowowowowow
@wesrothNew Claude model checkpoints have reportedly been detected in testing over the weekend, possibly connected to Anthropic’s broader Mythos rollout.
@wesrothAnthropic is reportedly planning to release a public version of Mythos tomorrow.
@ns123abcClaude 5 fable (mythos with stricter cyber guardrails) drops in the next 48 hours
@tolyI bet it won’t find any lof or dos bugs in percolator
@andrewcurran_
@cryptorover💥BREAKING: Antrophic is set to release Claude Mythos to the public by tomorrow!
@scaling01RT @claudeai: Introducing Claude Fable 5: a Mythos-class model that we’ve made safe for general use. Its capabilities exceed those of any…
@razrazcleRT @felixrieseberg: Today, we're introducing Claude Fable 5 and Mythos 5, two configurations of our next major language model. I'd normall…
@scaling01Fable 5 has the same underlying weights as Mythos 5 only with additional safeguards
@zephyr_z9"moving from giving AI tasks to giving it responsibilities."
@gavinsbakerEvidently prices will go up in about two weeks. Also I preferred Mythos to Fable. Fable makes me think of fabulist, which is not the best association. Excited to try it!
@hosseebHoly shit. Mythos just dropped (commercially restricted release called Fable), and these benchmarks are making every other lab look like amateurs. Really feels like it's Anthropic's world and we're all just living in it. Can't wait to see
@stocksavvyshayAnthropic announced Claude Fable 5 which is its new Mythos-class model for enterprise and paid users. Claude Fable 5 is also now available in $MSFT Foundry.
@hangsiinWhen Fable 5 is used for frontier LLM development, it does not notify the user and instead limits the model’s capabilities through methods such as prompt modification, steering vectors, and PEFT. Anthropic estimated that this would affect
@dair_aiRT @omarsar0: NEW: Anthropic introduces Claude Fable 5, a Mythos-class model for general use. Beginning of a new class of frontier models.
@jayelmnopRT @claudeai: Introducing Claude Fable 5: a Mythos-class model that we’ve made safe for general use. Its capabilities exceed those of any…