Anthropic Rolls Out Claude Mythos 5 AI Model—Along With the Safer Fable 5 for the Public
Anthropic launched Claude Mythos 5, a restricted-access model focused on advanced cybersecurity, and Claude Fable 5, a safer version for general use. Mythos 5 is initially limited to approved cybersecurity groups, critical infrastructure operators, government partners, and selected life sciences researchers because Anthropic says the model class now poses significant misuse risks. The company has been testing and limiting access through Project Glasswing while refining safeguards. The launch follows reports that earlier Mythos previews could autonomously complete complex intrusion tasks and find many browser vulnerabilities. Anthropic says Mythos 5 and Fable 5 score highly on cybersecurity, coding, biology, and reasoning benchmarks, and can handle longer, more complex tasks than prior Claude systems. To reduce abuse, sensitive requests about cybersecurity, biology, chemistry, or model replication in Fable 5 are routed to Claude Opus 4.8. Anthropic also plans to expand trusted access to Mythos 5 in consultation with the U.S. government.
