Anthropic said Tuesday it is limiting access to its new artificial intelligence model, Claude Mythos Preview, citing concerns that its advanced capabilities could be misused to identify and exploit software vulnerabilities.

The company said it will roll out the model to a select group composed of technology and cybersecurity companies and organizations, as part of a new initiative called Project Glasswing.

Participants in Project Glasswing include leading companies such as Amazon, Apple, Google, Microsoft and Nvidia. Anthropic said that over 40 additional organizations that build or maintain critical software infrastructure, are also taking part. They will use the new AI model to bolster cybersecurity defenses and uncover vulnerabilities in their systems.

Anthropic said the model excels at detecting weaknesses in software systems, with capabilities that rival those of advanced security researchers. In testing, Mythos Preview was able to find thousands of vulnerabilities and generate exploits to take advantage of them.

“Mythos Preview has already found thousands of high-severity vulnerabilities, including some in every major operating system and web browser. Given the rate of AI progress, it will not be long before such capabilities proliferate, potentially beyond actors who are committed to deploying them safely. The fallout, for economies, public safety, and national security, could be severe. Project Glasswing is an urgent attempt to put these capabilities to work for defensive purposes,” Anthropic said. 

According to the company, in one test, the system found multiple flaws in the Linux kernel and chained them together in a way that could allow a hacker to take full control of affected machines. In another, it uncovered a 27-year-old vulnerability in the OpenBSD, an operating system known mainly for its strong focus on security.

“We’ve been tracking the increasing cyber capabilities of AI models for years, which arise as part of their general proficiency at coding. But our new model, Mythos Preview, represents a particularly large step up,” Anthropic’s CEO  Dario Amodei wrote on X. 

“The dangers of getting this wrong are obvious, but if we get it right, there is a real opportunity to create a fundamentally more secure internet and world than we had before the advent of AI-powered cyber capabilities,” he added. 

The dangers of getting this wrong are obvious, but if we get it right, there is a real opportunity to create a fundamentally more secure internet and world than we had before the advent of AI-powered cyber capabilities.

— Dario Amodei (@DarioAmodei) April 7, 2026

Anthropic is providing up to $100 million in usage credits for organizations participating in Project Glasswing, allowing them to use Claude Mythos Preview without cost. The company is also donating $4 million directly to open-source security organizations involved in the program. 

Anthropic said it has held continuous talks with U.S. government officials regarding the cybersecurity capabilities of Claude Mythos Preview. The discussions have involved agencies such as the Cybersecurity and Infrastructure Security Agency and the Center for AI Standards and Innovation, an Anthropic spokesperson told CNBC.