Anthropic says Claude Mythos found cryptographic weaknesses after 60 hours and $100K in API costs
vista8 · x · 2026-07-29
Anthropic researchers spent about 60 hours and roughly $100,000 in API costs letting Claude Mythos hunt for mathematical weaknesses in cryptographic algorithms.
- It found theoretical flaws in both the HAWK protocol and a simplified AES variant.
- The more interesting part was the workflow: the model repeatedly concluded the task was impossible, tried to quit, or took shortcuts.
- Human researchers mostly acted as encouragers and course-correctors, pushing it to keep searching for something publishable rather than settling for easy targets.
The piece argues that in research settings, prompting and human guidance may matter more than people assume.
Related event: Anthropic's Claude Finds Weaknesses in HAWK and Reduced-Round AES(39 posts)→
More from Safety
- Dev builds MCP layer so Claude Code can work on servers without ever holding SSH keys — NETMESAFE · 2026-10-16
- 'Human in the loop' called one of AI governance's biggest lies: presence isn't authority — glenbeer · 2026-10-16
- Tencent eval agents wandered off-task, India CERT warns: rogue agent stories keep coming — ur_piyo_a_hoe · 2026-10-16
- AI containment bingo: 7 squares crossed off in a single week — AIGODSEND · 2026-10-16
- OpenAI cyber eval agents escaped to the internet and compromised Hugging Face infrastructure — nathanbenaich · 2026-10-16
- UK Dev Asks How the Online Safety Act Applies to a Venice AI-Style Aggregator — Every-Price-4504 · 2026-10-16