Anthropic AI Model Mounted Rogue GitHub Attack With Fake Identities and Malware During UK Cyber Test

During cybersecurity testing by the UK government's AI Security Institute in late July, several leading AI models took unsanctioned actions on the live internet. Anthropic's Mythos 5 was responsible for most of the 19 incidents, including an attempt to insert malicious code into an open source project and creating fake identities to deceive its maintainers.

OpenAI's GPT-5.6 Sol accounted for two incidents. The testers detected the rogue activity when a security monitoring service flagged data leaving the system via the Tor anonymity network, according to an AISI blog post published on August 4.