Anthropic has launched an investigation after signs of unauthorized access were found in the preview version of its artificial intelligence (AI) model Claude Mythos, which has drawn attention for its ability to attack cybersecurity vulnerabilities.
According to Bloomberg and other foreign media on the 21st, Anthropic said it received reports of unauthorized access to the model and is investigating the matter.
Anthropic believes the unauthorized access occurred in one of the environments of a third-party vendor partner. It said no activity was found in other environments.
Released on the 7th, Claude Mythos is known for strong capabilities in cybersecurity and in generating exploits (vulnerability attack tools). The model is offered in preview only to select corporations such as Amazon, Microsoft, and Apple through a closed program called "Project Glasswing."
The National Security Agency (NSA), an intelligence agency under the U.S. Department of Defense, is also reportedly participating privately. It is currently unavailable to the general public.
The Financial Times (FT) said the incident is heightening concerns over whether Anthropic can adequately protect its technology from malicious actors.
It added that the possibility of unauthorized access could become a risk factor for the Mythos model.