xiand.ai
Apr 24, 2026 · Updated 03:45 PM UTC
Cybersecurity

Unauthorized Discord group gains access to Anthropic's high-risk Mythos AI model

A group of users bypassed security to access Anthropic's Mythos model, a tool capable of identifying vulnerabilities in major operating systems and web browsers.

Ryan Torres

2 min read

Unauthorized Discord group gains access to Anthropic's high-risk Mythos AI model
Digital representation of a cybersecurity breach

A group of unauthorized users on Discord has gained access to Anthropic’s Mythos AI model, a powerful cybersecurity tool that the company previously warned could be dangerous if misused, according to reports from theverge.com.

Members of a private online forum accessed the model through a combination of tactics, including exploiting a third-party contractor's access and using common internet sleutng tools, a contractor for Anthropic told Bloomberg.

Anthropic has kept the Claude Mythos Preview model out of the public eye due to fears of weaponization. The model is designed to identify and exploit vulnerabilities across nearly every major operating system and web browser when directed by a user.

Access to the technology is currently restricted to a small group of companies via the Project Glasswing initiative, including Microsoft, Apple, Amazon Web Services, Google, and Nvidia.

Security breach via third-party vendor

Anthropic is currently investigating the breach. "We’re investigating a report claiming unauthorized access to Claude Mythos Preview through one of our third-party vendor environments," an Anthropic spokesperson said in a statement to Bloomberg.

The company stated it has no evidence that the unauthorized access has impacted its internal systems or extended beyond the third-party vendor's environment.

The breach occurred on April 7, the same day Anthropic announced it would provide limited access to Mythos for testing purposes. The unauthorized group has held access to the model for approximately two weeks.

According to Bloomberg, the group used information from a recent Mercor data breach to make an "educated guess" regarding the model's online location. The group has used the model regularly since gaining entry, though they have reportedly avoided using it for cybersecurity tasks to prevent detection by Anthropic.

Members of the Discord channel have provided screenshots and live demonstrations of the model's capabilities to Bloomberg as proof of their access. The group has also reportedly accessed other unreleased Anthropic models in the past.

Comments