Anthropic Rolls Out Claude Fable 5 With New AI Safeguards

Anthropic has begun rolling out Claude Fable 5, a new AI model built on the same foundation as its powerful Mythos class. When Anthropic first unveiled Mythos, the company warned that the frontier model was capable enough to help threat actors identify and exploit vulnerabilities in widely deployed software such as Firefox. In response, Anthropic restricted Mythos-tier models to vetted cybersecurity researchers and trusted enterprise partners to prevent offensive misuse.

Fable 5 represents Anthropic's attempt to broaden access to that capability tier while retaining strict safety guardrails. According to Anthropic, the model automatically blocks or redirects sensitive queries covering offensive cybersecurity, biology, and chemistry to its predecessor, Opus 4.8. An unrestricted counterpart, Claude Mythos 5, remains available only to a tightly controlled group that includes government cyberdefenders and specific life sciences researchers. Organizations concerned about their own exposure can validate their perimeter using a port scanner or check that certificates are properly configured with an SSL/TLS checker before deploying any new AI-driven tooling.

Until June 22, 2026, Fable 5 is free for Pro, Max, and Enterprise subscribers, after which Anthropic will switch to usage-based pricing due to the model's heavy compute requirements. Testing by BleepingComputer confirmed that Fable 5 burns through tokens at an extraordinary rate, consuming roughly 1 million tokens in 8 minutes and draining a fresh $100 Max subscription in as little as 9 minutes when paired with the new Workflow execution engine that spins up parallel subagents. Security teams experimenting with the model should audit their API keys and credentials via a password checker to ensure the rapid token usage does not translate into exposed secrets or accidental data leakage across development environments.