Fact Check

Anthropic's Claude Fable 5 Brings Mythos Power to the Public

TechCrunch · Tuesday, June 9, 2026 · Category: Models
Claim
Anthropic's Claude Fable 5 Brings Mythos Power to the Public

Anthropic released Claude Fable 5 to the general public on June 9, marking the first time its most powerful Mythos model has been made available outside of a closed partner program. The company said Fable 5 outperforms its predecessors in software engineering, knowledge work, and vision tasks, but it ships with hard safety guardrails: in high-risk domains including cybersecurity, biology, chemistry, and model distillation, the system refuses to answer and falls back to the older Claude Opus 4.8. Mythos itself was first previewed in April to a small group of partners citing cybersecurity concerns, and Anthropic expanded access last week to hundreds of organizations across 15 countries working on critical infrastructure. The public version is now reachable through Anthropic's Claude API and consumption-based Enterprise plans. Between now and June 22, Fable 5 is being bundled at no extra cost into Pro, Max, Team, and seat-based Enterprise subscriptions. After June 23, the model will be pulled from those tiers and billed against usage credits, with Anthropic saying it intends to return Fable 5 to standard subscriptions as soon as it can. In parallel, the company is rolling out Mythos 5 — the next iteration of the underlying flagship system — to organizations that have already cleared its approval process. The launch lands as Anthropic prepares for a public market debut alongside OpenAI and Elon Musk's SpaceX, and shortly after the company publicly called on major global AI labs to agree on a coordinated "brake pedal" for frontier development, warning that systems are advancing fast enough that recursive self-improvement — autonomous self-enhancement without human input — may soon be within reach. Anthropic disclosed that it stress-tested the model's classifiers before release, running an internal bug bounty that produced no universal jailbreaks across more than 1,000 hours of attempts, followed by external red-teaming engagements that also failed to crack the system. The company acknowledged, however, that novel attacks could still emerge. To address

View Original Source → Read Full Article →

← Back to News
Trending Topics
AICryptoBitcoinEthereumTechProgrammingStartupsWeb3DeFiNFTMachine LearningRoboticsCybersecurityCloud ComputingOpen SourceGamingFintechHealthTechEdTechClimate Tech