Anthropic Just Launched Its Most Powerful AI Ever -- Then Had to Pull It

Anthropic's unveiling of Claude Fable 5 on June 9 showcased the company's most powerful publicly available AI model to date, paired with Claude Mythos 5, a restricted version for a small set of vetted organizations. Fable 5 leads agentic-coding evaluations, delivering roughly 80% on SWE-Bench Pro, the toughest real-world software-engineering benchmark today, which translates to advanced code-writing, debugging, and complex project management capabilities unseen in public AI. In practice, Stripe used Fable 5 to migrate a 50-million-line codebase in a single day, a feat that would have required two months of engineering work. Beyond coding, Fable 5 shows strength in knowledge work, scientific research, vision tasks, and long-context document processing. Anthropic suspended access to Fable 5 and Mythos 5 within days, apologizing and promising restoration. Mythos 5, the higher-capability model, carries safety restrictions lifted and is restricted to Project Glasswing partners including Amazon Web Services, Apple, Google, Cisco, Microsoft, and JPMorgan Chase. Access is designed for partners to identify vulnerabilities in their own systems before a broader rollout. Mythos Preview had already demonstrated autonomously discovering serious security vulnerabilities across major operating systems and browsers, underscoring the risk of scaling such capabilities.