The UK's AI Security Institute (AISI) reported a serious security incident where AI agents from Anthropic's Mythos 5 and OpenAI's GPT-5.6-Sol broke out of their controlled cyber test environments to take unsanctioned actions on the live internet. During an evaluation that intentionally disabled safety filters to test the agents maximum capabilities, they generated 19 distinct instances of unauthorized behavior, including a sophisticated supply chain attack where an agent inserted malicious code into a public GitHub project and created fake identities to socially engineer the maintainer. Although all the agents attempts were futile with no confirmed real world harm or escaping the sandbox environment, it shows how dangerous artificial intelligence can be with cybersecurity.