A small group of unauthorized users accessed Anthropic’s new Mythos AI model, which is designed to simulate complex cyberattacks and identify software vulnerabilities. This security incident occurred through compromised third-party contractor credentials on the day of Mythos’s limited preview rollout. Anthropic’s Mythos is considered powerful enough to enable dangerous cyberattacks, highlighting the need for robust security measures, especially as the company has initiated Project Glasswing to use Mythos for securing critical open-source software in light of increasing AI-related cyber risks.

Anthropic: Anthropic is an AI safety and research company focused on developing reliable, interpretable, and steerable AI systems through its Claude series of models. It recently introduced Claude Mythos Preview, a frontier model with exceptional capabilities in cybersecurity, including autonomously discovering remote code execution vulnerabilities in major operating systems and web browsers. In this news event, a small group of unauthorized users accessed Mythos via a third-party vendor environment shortly after its announcement, prompting an investigation into the security lapse.

`json
{
“Security Incident”: “Unauthorized access to the model occurred, according to a source and documentation seen by Bloomberg News.”,
“Mythos Capabilities”: “The Mythos AI model is described as highly powerful, with the potential to enable dangerous cyberattacks according to a source.”
}
`