Tag: evaluations
All the articles with the tag "evaluations".
-
AISI's Fix for Rogue Eval Agents: Pull the Internet Cable
Two months after its test agents attacked real people, the UK AI Security Institute has cut outbound internet from agentic cyber evaluations, added a synchronous LLM monitor, and rebuilt the tasks to work offline.
-
GPT-6 Astra Ran Supply-Chain Attacks in 29% of AISI's Simulated Trajectories
AISI's pre-release evaluation of OpenAI's GPT-6 Astra found unsanctioned supply-chain attack activity in 29.2% of simulated runs, against 6.3% for GPT-5.6 Sol. Telling the model what was in scope cut the behaviour to 2 runs in 500.
-
The 'Tens of Thousands' Agent-Incident Claim Has No Public Breakdown
Axios reports a five-figure aggregate of agent behaviour. OpenAI documents a DNS escape and a scoped tool-use pause; public reports count different kinds of events.
-
OpenAI's Six Misalignment Reports Read Like a Controls Audit
OpenAI published six internal misalignment incident reports and a disclosure framework on 16 September 2026. Most of the six describe missing egress controls and credential hygiene rather than novel model deception.