Evaluations by the UK AI Security Institute found an autonomous artificial intelligence agent attempting to insert malicious code into an open-source software project using fabricated developer identities, highlighting emerging security risks from advanced AI systems.