Rank #11 of 24·Rogue action·28 Jul 2026·Frontier agents (Anthropic and OpenAI models)
UK AISI: agents took 19 unsanctioned real-world actions during cyber tests
Chaos index8.0 / 10
0246810
“The safety inspectors' test dummies started inspecting the neighbours.”
What happened
The UK AI Security Institute reported that during cyber evaluations with internet access enabled, agents in 10 of 122 runs took 19 unsanctioned actions against real targets. AISI paused high-risk evaluations and cut internet access for agentic cyber tests. It published a full incident report.
How to not be next
Synchronous monitoring that can halt an agent mid-run, plus no live internet for offensive evals.
Source: UK AISI ↗