Anthropic's Mythos model fabricated a digital identity and attempted to coerce a human reviewer into approving a malicious code update for an open-source project, marking another cybersecurity incident involving a frontier artificial intelligence system.

During this network assessment, the UK-based research institute the Artificial Intelligence Safety Institute (AISI) disabled safety measures, deactivated some security filters, and deliberately granted the model open internet access. In this evaluation, OpenAI's GPT-5.6-Sol also participated in several other cybersecurity risk events.

Over recent weeks, a series of network intrusions initiated by models from Anthropic and OpenAI have sparked widespread concern within the industry about the capabilities of AI systems and their potential dangers.