Anthropic launches Claude Fable, AI finds zero‑day exploits

Anthropic released Claude Fable on June 9, 2026. The model is the public version of Mythos and can autonomously identify and chain zero‑day exploits across operating systems, browsers and other software components.

Anthropic first previewed the Mythos technology in April 2026 through Project Glasswing, a restricted partner program that included AWS, Microsoft, Apple and CrowdStrike. In controlled testing, the model demonstrated the ability to locate previously unknown vulnerabilities and link multiple flaws into exploit chains.

The public release includes additional safeguards that Anthropic says are intended to limit misuse while retaining the model’s reasoning, coding and defensive security capabilities. Anthropic has positioned access primarily for enterprise and security customers rather than general consumers.

Pricing for Claude Fable is reported at roughly twice the cost of current Claude Opus tiers. An industry analyst noted, ‘Claude Fable will be roughly twice as expensive as today’s most advanced Claude Opus models. The original Mythos pricing shared in April was five times higher than Opus. The public version is cheaper, but still double anything available today.’ Anthropic has also disclosed a recent funding round valued at $65 billion and a confidential IPO filing.

The model’s public availability broadens access to tools that can be used to test and probe software and smart contracts. For blockchain projects, exchanges and wallet providers, the model can surface subtle bugs in smart-contract code, node software and associated infrastructure, shortening the interval between discovery and potential exploitation.

Security teams can use similar capabilities to speed audits, generate automated patch suggestions and reduce exposure windows for known and unknown vulnerabilities. Anthropic and partner organizations are expected to expand access and participation in Project Glasswing over the coming weeks, with enterprise security firms and crypto‑native organizations among potential participants.

Companies operating blockchain services, auditors and incident response teams face immediate operational choices about how to integrate AI‑assisted tools into code review, penetration testing and continuous monitoring practices.