A small group of unauthorized users gained access to Anthropic's new AI model, Mythos. This was reported by Bloomberg, citing internal documents.
According to the agency, several members of a closed online forum were able to use the neural network on the day it was released and have since been using it regularly.
Anthropic positions Mythos as a system capable of detecting and exploiting vulnerabilities in "all major operating systems and web browsers." As a result, the company has restricted access to it to a select group of software providers.
To breach the system, users employed several tactics: they used the credentials of an employee from a third-party contractor of Anthropic, guessed the model's URL based on the addressing patterns of other company systems, and extracted additional information from a data leak at the startup Mercor.
A Bloomberg source claims that the group intends to experiment with the new model and does not plan to cause harm. In addition to Mythos, its members also have access to several other unreleased neural networks from Anthropic.
"We are investigating reports of unauthorized access to the Claude Mythos Preview through one of the environments of third-party providers," said a company representative.
This incident highlights the challenges of controlling the spread of potentially dangerous technologies and raises the question of who else might gain access to Mythos and for what purposes.
Mythos's Capabilities
Mozilla reported in its blog that an early version of Mythos helped identify 271 vulnerabilities in the Firefox browser during internal testing. The errors have since been fixed.
The results demonstrated how advanced AI systems can analyze large codebases and find weaknesses that previously required thorough examination by cybersecurity experts.
Earlier, Mozilla tested another Anthropic model that identified 22 vulnerabilities in a previous version of Firefox. Despite the new findings, the company acknowledges that achieving absolute security is an "unrealistic goal."
The firm stated that all identified errors could have been found by a highly skilled human researcher.
"Some commentators believe that future AI models will discover entirely new forms of vulnerabilities that go beyond our current understanding. We do not share that view," the company noted.
It is worth mentioning that in April, media outlets reported that the U.S. National Security Agency is using Mythos despite the startup's conflict with the Pentagon.
In March, the company confirmed a leak of part of the source code of its AI programming tool, Claude Code.
Later, it accidentally deleted thousands of repositories on GitHub in an attempt to rectify the situation.