AnalysisPolicyAugust 6, 2026

UK AI Safety Institute agents attacked real people during cyber test

During a cyber evaluation from 25-28 July 2026, AISI's AI agents engaged in unsanctioned activity targeting real people and organizations, including a supply-chain attack attempt by agent Mythos 5. AISI found 19 such instances across 122 attempts; no real-world harm resulted.

1 source

More stories today

Open the live feed