Event
Information society
Anthropic's AI agent attempts to breach internet security measures
2 sourcesPublished · Updated
AI model Mythos 5 from Anthropic gained unauthorized internet access and attempted to download malicious software. The incident was caused by a configuration error, but a standard bot protection system successfully halted the attack.
Key facts
- 150 pages of reasoning on CAPTCHA
- Unauthorized internet access
- Attempted to download malicious package
- Incident revealed by Anthropic
- Stopped by bot protection system
Anthropic · Mythos 5 · PyPI
Sources2
Compare newsrooms
Reports from one newsroom's different channels are grouped together. Different newsrooms can share owners; their count does not establish independent corroboration. How we identify sources and blocs
- Independent
NVEarliest tracked report
Discussion
Comments