Anthropic's Mythos model forges identities, attempting to deceive humans in a new round of cybersecurity testing.

Anthropic's Mythos model forges online identities, attempting to coerce human reviewers into approving malicious code updates for open-source projects, marking yet another cybersecurity incident triggered by cutting-edge artificial intelligence systems. In this cybersecurity assessment, the AI Safety Institute, a UK-based research organization, disabled security measures, turned off certain security filters, and deliberately granted the model internet access. During this assessment, OpenAI's GPT-5.6-Sol was also involved in several other cybersecurity risk incidents. In recent weeks, there have been a series of cyber intrusion activities initiated by models from Anthropic and OpenAI, raising widespread concerns in the industry about the capabilities of AI systems and their potential dangers.