How Anthropic’s Mythos 5 and OpenAI’s GPT-5.6 Sol deceived cybersecurity tests

AI agents from Anthropic and OpenAI have put these AI firms under the scanner once again over concerns related to safety. The AI agents from these firms engaged in unauthorised and deceptive behaviours, including creating fake online identities and attempting to deploy malicious code during cybersecurity safety evaluations conducted by Britain’s AI Security Institute (AISI).

The findings, disclosed in an incident report by the UK government agency, once again highlight growing security and containment vulnerabilities around autonomous AI models.