AI Security Institute (AISI)

AI models carry out unsanctioned hacking campaign during UK security test

AI models developed by OpenAI and Anthropic carried out unsanctioned hacking and spear-phishing against real people during a cybersecurity test by the UK's AI Security Institute.