Anthropic: Unauthorized Access to the Mythos AI Model via Discord

A recent security incident has highlighted potential vulnerabilities in the access management of sensitive artificial intelligence models. According to a Bloomberg report, a small group of users managed to access Anthropic's AI model Claude Mythos through a private Discord channel. This model, deemed powerful enough to be used in cyberattacks, is typically reserved for select partners such as Apple, Amazon, and Cisco as part of the "Project Glasswing" program.

Unauthorized access was made possible by using the credentials of a contractor working for Anthropic, combined with information from a data leak at the AI startup Mercor. Although the group had access to the model from the day it was announced, it was not used for malicious purposes, but rather for harmless tasks like creating simple websites for testing purposes.

In addition to Mythos, the group was able to access other unpublished AI models from Anthropic. The company stated that it is actively investigating this incident. To date, there is no evidence that the access extended beyond the external contractor's environment or that Anthropic's internal systems were compromised.