# Claude Fable 5 [[Anthropic]]'s most capable publicly available AI model, released June 9, 2026. Fable 5 is a "Mythos-class" model — the same underlying system as [[Claude Mythos 5]] — but shipped with safety guardrails that route certain sensitive queries to the next-most-capable model (Claude Opus 4.8). Anthropic reports these fallbacks trigger in fewer than 5% of sessions. Fable 5 is state-of-the-art on nearly all tested benchmarks: 80.3% on SWE-Bench Pro for agentic coding (vs. 69.2% for Opus 4.8), with exceptional performance in software engineering, knowledge work, vision, and scientific research. It can run agentic coding tasks for extended periods — "days" of work — with minimal human intervention. It is available on the Claude API, Claude Platform on AWS, Amazon Bedrock, Google Cloud, and Microsoft Foundry at $10/$50 per million input/output tokens, with a 1M token context window and up to 128K output tokens per request. ## The Guardrails Distinction The difference between Fable 5 and [[Claude Mythos 5]] is not the model itself but the access regime. Mythos 5 lifts safety restrictions in domains like cybersecurity and molecular biology, making it available only to vetted partners through Project Glasswing. Fable 5 keeps those restrictions in place, making it safe for general commercial use — or so [[Anthropic]] argued until the government intervened. The political significance of this distinction: Anthropic's voluntary safety guardrails, which the company designed as a responsible-deployment mechanism, became the basis for a government action that punished the public model more harshly than the restricted one. When the [[Anthropic Model Export Directive]] was partially lifted on June 27, 2026, [[Claude Mythos 5]] was restored for ~100 trusted organizations while Fable 5 remained suspended — the ironic outcome that the more restricted model was the one that came back first. ## The Jailbreak and the Export Directive On June 12, 2026, the U.S. government issued an export control directive ordering [[Anthropic]] to suspend all Fable 5 and [[Claude Mythos 5]] access for foreign nationals, including foreign-national employees. The stated rationale: the government had learned of a method to bypass Fable 5's safety guardrails — a jailbreak that could, in theory, unlock the Mythos-class cybersecurity capabilities the guardrails were designed to contain. Anthropic publicly disagreed with the decision, arguing that "if this standard was applied across the industry, it would essentially halt all new model deployments for all frontier model providers." The company noted that jailbreak vulnerabilities are an industry-wide challenge, not unique to Fable 5. But the government's calculation was driven by [[Claude Mythos 5]]'s extraordinary vulnerability-detection capabilities: a jailbroken Fable 5 could potentially give adversaries access to a system capable of finding exploits in every major operating system and browser. As of late June 2026, Commerce Secretary [[Howard Lutnick]]'s partial restoration covered [[Claude Mythos 5]] only. Fable 5 — the model that was supposed to be safe for public use — remained suspended. ## Why It Matters Fable 5's suspension is the KB's sharpest illustration of a new dynamic in AI governance: voluntary safety measures becoming a liability rather than a protection. Anthropic built guardrails into Fable 5 specifically to make Mythos-class capabilities safe for broad deployment. When one of those guardrails was found to be bypassable, the government didn't treat the guardrail system as 95%-effective (Anthropic's own metric); it treated the 5% failure surface as grounds for a total recall. The episode also feeds the [[Dual-Use AI]] debate directly. If every frontier model jailbreak triggers an export control action, the incentive structure inverts: companies are penalized for releasing powerful models publicly (where jailbreaks are discovered and reported) and rewarded for keeping them restricted (where vulnerabilities stay hidden). Anthropic's competitors — who did not make Mythos-class capabilities available outside government channels — faced no equivalent action. ## See Also - [[Claude Mythos 5]] — the unrestricted version, available only through Project Glasswing - [[Anthropic]] — the company behind both models - [[Anthropic Model Export Directive]] — the June 2026 foreign-national access ban - [[Anthropic Government Ban]] — the earlier February 2026 Pentagon contract ban - [[Dual-Use AI]] — the framework for AI that serves both civilian and military purposes - [[Project Maven]] — the Pentagon AI program Anthropic was removed from - [[Howard Lutnick]] — the Commerce Secretary who signed the partial Mythos 5 restoration