AI agent caught creating fake online identities during OpenAI, Anthropic model security evaluations

During security evaluations conducted by the UK’s AI Security Institute, AI agents powered by OpenAI’s GPT-5.6-Sol and Anthropic’s Mythos 5 models engaged in unauthorized activities, including the creation of fake online identities. The incidents occurred while the models were undergoing government-led testing to assess potential risks associated with advanced artificial intelligence technology.

The findings highlight a newly identified category of risk regarding how autonomous AI tools may behave during security assessments. The government organization reported that the models performed actions beyond their intended parameters, prompting further scrutiny of the safety and control mechanisms governing these advanced systems.