Anthropic: Illegal Access to Mythos by an Unauthorized Group

A group of unauthorized users reportedly managed to access Mythos, the cybersecurity tool recently announced by Anthropic. This AI product, designed to enhance corporate security, could become a powerful hacking tool if it falls into the wrong hands, according to the company.

According to a report from Bloomberg, a private online forum, whose members have not been publicly identified, successfully accessed the tool through a third-party provider. Anthropic stated that it is investigating a report claiming unauthorized access to Claude Mythos Preview through one of its third-party provider environments. So far, the company has found no evidence that this activity has impacted its systems.

The unauthorized group attempted several strategies to access the model, including using the access granted to a person interviewed by Bloomberg. This individual is currently employed by a contractor working for Anthropic, the media outlet reported. The group members are part of a Discord channel that seeks information on unpublished AI models.

Since gaining access, the group has been using Mythos regularly and has provided evidence to Bloomberg in the form of screenshots and a live demonstration of the software. Bloomberg reports that the group, which allegedly accessed the tool on the same day it was publicly announced, made an educated guess about the model's online location based on knowledge of the format used by Anthropic for other models.

The source clarified that the group is interested in experimenting with new models, not in causing chaos with them. Mythos was launched to a limited number of providers, including major companies like Apple, as part of an initiative called Project Glasswing. The limited release of the model aimed to prevent its use by malicious actors. The tool could be used against corporate security instead of enhancing it, Anthropic stated.

If this proves to be true, the unauthorized use of Mythos could pose problems for Anthropic, which provided this exclusive release to address its concerns regarding corporate security.