Anthropic Withholds Release of Powerful AI Model Mythos for Cybersecurity Coalition

Anthropic revealed on Tuesday a new AI model named Claude Mythos, described as highly powerful, and decided against a public release to mitigate cybersecurity risks. The model has identified thousands of bugs and vulnerabilities in popular software, including every major operating system and web browser.

Instead of broad availability, Anthropic initiated Project Glasswing, a collaborative effort with cybersecurity and technology companies to explore defensive applications.

Project Glasswing includes participants such as Microsoft, Amazon Web Services, Apple, Google, Nvidia, CrowdStrike, and Palo Alto Networks. The coalition aims to allow these organizations to flag vulnerabilities in their systems with minimal human intervention. Anthropic reported that the model preview has already uncovered significant security issues across critical software.

Coalition Details and Scope The initiative involves between 40 and 45 organizations, according to reports from multiple sources.

Wired specified more than 45 participants, while others cited 40 companies. This partnership focuses on advancing AI-driven cybersecurity capabilities to prevent cyberattacks. Anthropic emphasized that the model's capabilities could enable rapid detection of flaws in software environments.

The company is providing preview access to coalition members to test and refine these applications. No timeline for wider release was provided.

Model Capabilities and Risks Claude Mythos demonstrated its potential by finding security problems in all major operating systems and web browsers, as stated by Anthropic.

CoinDesk reported that it identified thousands of such issues in popular programs. The decision to limit access stems from concerns that hackers could exploit the model for cyberattacks.

“— CoinDesk, date not specified This controlled rollout prioritizes defensive uses over general deployment. Anthropic's approach highlights ongoing debates in AI development regarding safety and dual-use technologies.”

Background on Anthropic's Strategy Anthropic, known for its Claude AI series, positions this project as a step toward responsible AI innovation. The New York Times described the announcement as a cybersecurity reckoning, though the company focused on practical testing. Coalition members will use the model to enhance their security protocols. The initiative could extend to government applications, as suggested by The Verge, but details remain limited. Anthropic's rivals, including key tech giants, are participating despite competitive dynamics. This collaboration underscores the industry's recognition of AI's role in both threats and defenses.