tech
What Happens When Anthropic's Mythos Class Models Go Public?
Claude Mythos 5, touted as the world’s most powerful AI cybersecurity model, launches via Project Glasswing as Fable 5 goes public
TL;DR
- Anthropic launches Claude Fable 5, a Mythos-class AI model safe for general use, and Claude Mythos 5, the most powerful AI cybersecurity model.
- Fable 5 can hold entire projects, plan, run for extended periods, and check its own work, significantly compressing task timelines.
- Mythos 5, released via Project Glasswing, possesses advanced cybersecurity capabilities with some safeguards lifted compared to Fable 5.
- Safeguards redirect certain requests from Fable 5 to Claude Opus 4.8 to prevent dangerous use cases, though this may occasionally impact harmless requests.
- Mythos 5 can perform agentic hacking across multiple stages and has undergone extensive red teaming with no successful jailbreaks reported.
- Biological research restrictions have been expanded due to concerns about 'well-resourced malicious actors' using models for risky research, with Fable 5 falling back to Opus 4.8 for most biology and chemistry requests.
- Industry reactions vary, with some questioning the adequacy of safeguards against the inherent risks of such powerful AI models.