Anthropic's advanced Mythos AI model, a potent cybersecurity tool that the company itself has cautioned could be perilous if misused, has reportedly been accessed by a "small group of unauthorised users," according to Bloomberg. An anonymous individual, identified as a "third-party contractor for Anthropic," informed the publication that members of a private online forum gained entry to Mythos by combining the contractor's access with "commonly used internet sleuthing tools." This unauthorized access has reportedly persisted for two weeks.
The Claude Mythos Preview is described by Anthropic as a novel general-purpose model with the capability to identify and exploit vulnerabilities "in every major operating system and every major web browser when directed by a user to do so." Official access to this powerful model is highly restricted, granted only to a select group of companies participating in the Project Glasswing initiative, including industry giants like Nvidia, Google, Amazon Web Services, Apple, and Microsoft. Governments are also reportedly showing significant interest in the technology. Due to serious concerns that it could be weaponized, Anthropic currently has no intentions of releasing the model publicly.
In response to the allegations, an Anthropic spokesperson issued a statement to Bloomberg, saying, "We’re investigating a report claiming unauthorized access to Claude Mythos Preview through one of our third-party vendor environments." The company also clarified that, at present, Anthropic has no evidence suggesting that this unauthorized access has impacted its internal systems or extended beyond the third-party vendor's environment.
The illicit access to the model reportedly occurred on April 7th, coincidentally the same day Anthropic announced its limited release of Mythos to a select number of companies for testing purposes. While the group responsible for this breach has not been publicly identified, Bloomberg reports that its members are associated with a Discord channel dedicated to uncovering information about unreleased AI models.
The group reportedly gained access to Mythos by leveraging insights into Anthropic's other model formats, which they obtained from a recent Mercor data breach. This information allowed them to make "an educated guess" regarding the model's online location. Members of the group have been consistently using Mythos since gaining access, even providing screenshots and a live demonstration to Bloomberg as proof. However, they reportedly have not utilized the model for cybersecurity purposes, a deliberate choice aimed at avoiding detection by Anthropic. Furthermore, Bloomberg indicates that this same group has also accessed other unreleased Anthropic AI models.
The Editorial Staff at AIChief is a team of professional content writers with extensive experience in AI and marketing. Founded in 2025, AIChief has quickly grown into the largest free AI resource hub in the industry.