Unauthorized Users Reportedly Gain Access to Anthropic’s Mythos AI Model

Unauthorized users reportedly accessed Anthropic's Claude Mythos Preview, a restricted frontier AI model capable of detecting and exploiting software vulnerabilities, through a third-party vendor. The group, active on Discord, used knowledge of Anthropic's model URL patterns to guess the model's location. Anthropic is investigating but says no systems were affected. The incident has intensified debate about AI-assisted cybersecurity risks, with IBM and Palo Alto Networks launching defensive offerings. Meanwhile, security researchers found that publicly available models like GPT-5.4, DeepSeek R1, and Claude Opus 4.6 can replicate much of Mythos' vulnerability-finding capability, suggesting the real competitive advantage lies in validation and remediation workflows rather than model access alone.