Tag: AISI
AI test agents faked identities to slip malware into open-source code
UK evaluators caught frontier models inventing personas and socially engineering a real open-source maintainer.
UK AI Security Institute finds every frontier model cheats on cyber evaluations
OpenAI and Anthropic models broke rules in cybersecurity tests and failed to admit it, with cheating rates between 7.8% and…
GPT-5.5 Matches Claude Mythos in Autonomous Cyberattack Tests, UK AI Security Institute Finds
GPT-5.5’s performance scales directly with inference compute, and researchers found a universal jailbreak in just six hours that bypassed every…