News
AI · Anthropic · Cybersecurity · Mythos
US-based AI developer Anthropic announced its powerful AI model, Claude Mythos Preview, is too dangerous for public release due to its exceptional ability to find high-severity cybersecurity vulnerabilities in major operating systems and web browsers, leading the company to restrict access to a limited set of partners.
Anthropic stated on Tuesday that Mythos Preview is too effective at discovering weaknesses, which cybercriminals and spies could abuse. A March data leak first revealed Mythos's "unprecedented cybersecurity risks," causing cybersecurity stocks to slump.
The model demonstrated alarming capabilities, including bypassing virtual sandbox security and autonomously chaining Linux kernel errors to take complete control of machines. It also discovered a 27-year-old vulnerability in OpenBSD, an operating system heavily used in critical infrastructure.
Anthropic will make Mythos Preview available to 12 organizations, including Amazon Web Services, Apple, Broadcom, Cisco, CrowdStrike, Google, JPMorganChase, the Linux Foundation, Microsoft, Nvidia, and Palo Alto Networks, as part of "Project Glasswing" for defensive cybersecurity. Anthropic is in "ongoing discussions" with US government officials regarding Mythos's cyber capabilities, emphasizing the need for the US and its allies to maintain AI leadership.
CEO Dario Amodei expects similar powerful models from competitors within six to eighteen months, underscoring the security industry's need to prepare for these advanced capabilities.
