Anthropic’s Mythos AI model, a powerful cybersecurity tool that the company said could be dangerous in the wrong hands, has been accessed by “a small group of unauthorized users.” bloomberg Report. An unnamed member of the group, identified only as “a third-party contractor for Anthropic,” told the publication that members of a private online forum became involved in Mythos through a mix of tactics, including contractor access and using “commonly used Internet investigative tools.”
According to Anthropic, Cloud Mythos Preview is a new general-purpose model that is “capable of identifying and exploiting vulnerabilities in every major operating system and every major web browser, when directed to do so by the user”. Official access to the model through the Project Glasswing initiative is limited to a handful of companies, including Nvidia, Google, Amazon Web Services, Apple, and Microsoft. Governments also have an eye on technology. Anthropic currently has no plans to release the model publicly due to concerns that it could be weaponized.
“We are investigating a report claiming unauthorized access to Cloud Mythos Preview through one of our third-party vendor environments,” an Anthropic spokesperson said in a statement. bloomberg. Anthropic currently has no evidence that unauthorized access is affecting the company’s systems or beyond the third-party vendor’s environment.
The model was reportedly accessed illegally on April 7, the same day that Anthropic announced it was releasing Mythos to a limited number of companies for testing. However, the group that gained unauthorized access has not been publicly identified. bloomberg Its members are reportedly part of a Discord channel that seeks information about unreleased AI models.
The group accessed Mythos using knowledge of Anthropic’s other model formats obtained from the recent Mercor data breach to make “an educated guess” about its online location. Members have been using Mythos regularly since gaining access – providing screenshots and live demonstrations of the model as evidence bloomberg – although reportedly not for cyber security purposes in an effort to avoid detection by Anthropic. Other unpublished anthropic AI models have also been accessed by the group bloomberg.
