Anthropic Offers Mythos Upgrade for Cyber Partners and a ‘Safe’ Version for the Rest of You

Key Points:

• Anthropic launched two new AI models, Claude Fable 5 and Claude Mythos 5, with enhanced capabilities compared to its April Mythos Preview, but Mythos 5 remains limited to select industry partners and government collaboration due to cybersecurity concerns.
• Claude Fable 5 is publicly available but includes strict "guardrails" that block queries related to cybersecurity, biology, and chemistry, rerouting sensitive requests to an older model, Claude Opus 4.8, to prevent misuse such as hacking tool development or AI distillation.
• The company emphasizes caution in releasing these models broadly, acknowledging that some benign user queries may be restricted, with plans to improve classifier precision over time to balance safety and usability.
• Anthropic is providing unrestricted access to Mythos 5 for select biology researchers and Project Glasswing partners as part of a controlled rollout, while preparing for a future trusted access program to expand availability safely.
• The release reflects ongoing industry tension between advancing AI capabilities and addressing cybersecurity risks, with Anthropic and competitors like OpenAI working on safeguards amid investor pressures as they approach public offerings.