Key Highlights:

• The model remains restricted to select organizations for defensive use only.
• Anthropic previews its new frontier AI model, Mythos, for cybersecurity research.
• Project Glasswing includes partners like Apple, Microsoft, Amazon, and Cisco.
• Mythos reportedly detected thousands of zero-day vulnerabilities recently.

Anthropic has released a limited preview of its powerful new AI model Mythos as part of a cybersecurity collaboration called Project Glasswing. The initiative allows selected organizations to test the system for defensive security work and software vulnerability detection. Anthropic says the model has already identified thousands of zero-day flaws across software systems.

The announcement signals a new phase in how frontier AI models may support global cybersecurity infrastructure.

What Is Anthropic’s Mythos AI Model?

Anthropic describes Mythos as one of its most advanced frontier models designed for complex reasoning, coding, and agentic workflows. Frontier models represent the company’s highest-performance AI systems and typically power advanced Claude AI capabilities.

Although Mythos was not trained specifically for cybersecurity tasks, Anthropic says it performs strongly in vulnerability detection. Recently, the model scanned both internal and open-source software environments. As a result, it discovered thousands of zero-day vulnerabilities, many dating back one or two decades.

These findings suggest that legacy software risks remain widespread across critical digital systems.

What Is Project Glasswing and Who Is Involved?

Project Glasswing is Anthropic’s new cybersecurity initiative built around collaboration rather than public deployment. The program currently includes 12 partner organizations working together on defensive security applications.

Participants include Amazon, Apple, Broadcom, Cisco, CrowdStrike, the Linux Foundation, Microsoft, and Palo Alto Networks. These organizations are expected to share technical insights from their testing so the broader industry can benefit later.

However, Anthropic confirmed that Mythos will not be released for general public access at this stage. Instead, around 40 additional organizations beyond the core partners will receive preview access under controlled conditions.

Why Did Mythos Detect So Many Zero-Day Vulnerabilities?

Anthropic says Mythos identified thousands of hidden vulnerabilities during recent scans of software infrastructure. Many of these flaws had existed unnoticed for years.

Zero-day vulnerabilities are particularly critical because attackers can exploit them before developers release patches. Therefore, early detection can reduce large-scale security risks across industries.

The company also warned that frontier models like Mythos could become dangerous if misused. In leaked internal material, Anthropic reportedly noted that the same capability used for defensive scanning could also enable malicious exploitation.

How Did the Mythos Leak Reveal Its Capabilities Earlier?

News of Mythos first surfaced after a data exposure incident last month. A draft blog post describing the model appeared in a publicly accessible document cache.

At that time, the system carried the internal name “Capybara.” According to the leaked draft, Mythos exceeded performance benchmarks previously reached by Anthropic’s Opus models in coding, reasoning, and cybersecurity testing.

Anthropic later confirmed the exposure resulted from human error.

Are Government Discussions Affecting Mythos Deployment?

Anthropic said it has ongoing discussions with federal officials about the model’s future role in security environments. However, those conversations come during a legal dispute with the Pentagon.

The U.S. Department of Defense previously labeled Anthropic a supply-chain risk after the company declined to support autonomous targeting or surveillance involving American citizens. That disagreement continues to shape the broader policy environment around advanced AI deployment.

Still, Project Glasswing signals that Anthropic intends to position Mythos as a defensive cybersecurity tool rather than a public-facing product. As frontier AI systems evolve, Anthropic appears to be testing how controlled access models like Mythos can strengthen software security at scale.