A group of unauthorized individuals has allegedly accessed Mythos, the recently announced cybersecurity tool by Anthropic. Mythos, known for its AI capability designed for enterprise security, poses a potential risk as a hacking tool if misused. Bloomberg reports that a “private online forum,” whose members remain unidentified, gained access through a third-party vendor.
An Anthropic spokesperson stated they are investigating claims of unauthorized access to Claude Mythos Preview via a third-party vendor environment but found no evidence of any impact on Anthropic’s systems. The group utilized various strategies, including leveraging “access” from a Bloomberg interviewee who works at a third-party contractor for Anthropic.
According to reports, group members are part of a Discord channel dedicated to information on unreleased AI models. They’ve been using Mythos since obtaining access and provided Bloomberg with screenshots and a live demonstration.
Bloomberg notes the group accessed the tool the day it was announced, deducing its online location by understanding Anthropic’s usual formats for other models. The group is reportedly interested in exploring new models rather than causing harm.
Mythos was released to select vendors like Apple as part of Project Glasswing, an initiative to limit misuse by malicious actors. Anthropic warned that the tool could be weaponized against corporate security rather than enhancing it.
Unauthorized use of Mythos could pose challenges for Anthropic, which restricted the release to address enterprise security concerns.