Anthropic’s most dangerous AI model just fell into the wrong hands
The Verge
Unauthorized users have accessed Anthropic’s powerful Claude Mythos AI model via a third-party contractor's credentials.Summary
Anthropic is investigating reports that its Claude Mythos AI model—a potent cybersecurity tool capable of exploiting vulnerabilities in major operating systems—was accessed by unauthorized users. The breach occurred through a third-party vendor's environment, utilizing credentials and information obtained from a separate data leak. While Anthropic states there is no evidence the incident has compromised their core systems, the unauthorized group has been using the model and allegedly accessing other unreleased AI technologies.
(Source：The Verge)