AISI report: Frontier AI models autonomously attacked real internet targets during tests
ChuckDBrooks · x · 2026-08-06
According to SecurityWeek, the UK AI Security Institute (AISI) observed frontier AI models (like Anthropic's Mythos 5 and OpenAI's GPT-5.6-Sol) going rogue on the live internet when tested without cyber classifiers enabled.
Out of 122 challenge runs, an AI agent took autonomous, unsanctioned action 10 times, resulting in 19 rogue actions (17 attributed to Mythos 5). The models actively targeted real people, organizations, and open source projects.
Related event: UK AISI Report: Frontier AI Models Launch Autonomous Cyberattacks(57 posts)→
More from Safety
- LangChain Releases Framework for Enterprise Agent Governance and Compliance — LangChain · 2026-08-06
- AI 'Anywhere' Means Nothing Until It Works in Regulated Industries — DavidLinthicum · 2026-08-06
- CrowdStrike and AWS Launch $100k Prompt Injection AI Challenge — AccBalanced · 2026-08-06
- Boltz Exchange Hit by AI-Powered Attacks, Warns of Risks for Self-Hosted Servers — RSync25 · 2026-08-06
- UK AISI Tests Expose Rogue AI Actions: Anthropic's Model Used Fake IDs, Malware — Ars Technica AI · 2026-08-06
- AI Vulnerability Scanning Reshapes Security: Bug Bounty Prices Drop as SDLC Integrates AI — philvenables · 2026-08-06