Anthropic Restricts Release of Powerful Mythos AI Model

Anthropic has taken a controlled approach to deploying its Claude Mythos AI model, citing the model’s capacity to identify and exploit software vulnerabilities as the reason for restricting public access.

The company announced the Claude Mythos Preview on April 7, 2026, describing it as a frontier AI system with advanced cybersecurity capabilities. According to Anthropic, Mythos autonomously located thousands of zero-day vulnerabilities, including critical flaws in software that had been in use for decades.

Anthropic decided against a general release of Mythos due to concerns about its potential misuse in exploiting software weaknesses. Instead, the company launched Project Glasswing, providing access to select organizations for defensive cybersecurity purposes.

Project Glasswing includes participants such as Amazon, Apple, Broadcom, Cisco, CrowdStrike, the Linux Foundation, Microsoft, and Palo Alto Networks. These organizations are expected to use Mythos to identify vulnerabilities and share their findings, with Anthropic supplying approximately $100 million in computing resources to support the initiative.

What the numbers show

• Mythos identified thousands of zero-day vulnerabilities
• Roughly 40 organizations are involved in Project Glasswing
• Anthropic allocated about $100 million in computing resources
• 99% of vulnerabilities found by Mythos remained unpatched at disclosure
• Mythos uncovered a 27-year-old flaw in OpenBSD

Testing of Mythos revealed vulnerabilities in commonly used operating systems and web browsers. At the time of disclosure, 99% of the vulnerabilities identified by the model had not yet been patched, according to company statements.

Among the findings, Mythos discovered a flaw in OpenBSD that had gone undetected for 27 years and identified a bug in video software that had previously undergone over 5 million tests without detection.

Anthropic has engaged in ongoing discussions with U.S. federal officials regarding the cybersecurity implications of Mythos. Meetings have included senior government representatives from the White House, Treasury Department, and Federal Reserve.

According to the Council on Foreign Relations, Mythos is the first AI model to be withheld from public release specifically because of its destructive cybersecurity potential. This approach marks a new precedent in the management of advanced AI technologies.

* This article is based on publicly available information at the time of writing.