Claude Mythos, a new AI supermodel, identified more than ten thousand software vulnerabilities in just one month.

Anthropic Releases First Report on Cybersecurity Project “Project Glasswing”

Anthropic announced the results of its first public review of a cybersecurity initiative launched in April this year. The project centers around an AI model, Claude Mythos Preview, which is not yet available to the general public.

How the Model Works
In its first month of operation, Anthropic partners used Mythos Preview to scan their systems and identified over 10,000 software vulnerabilities. In most cases, the bugs were critical, enabling companies to respond quickly to threats.

Results Among Corporate Clients
Company | Number of Vulnerabilities Found | Critical/High-Priority
Cloudflare | 2,000 | 400
Mozilla | 271 (fixed in Firefox) |
Microsoft | Increase in patch volume linked to Mythos

- Cloudflare identified two thousand vulnerabilities, of which four hundred were classified as high or critical risk.
- Mozilla discovered and fixed 271 bugs in the Firefox browser, significantly surpassing the results from previous versions of the Claude model.
- Microsoft plans to expand its patch releases based on infrastructure scans conducted with Mythos Preview.

Open Source Projects
Anthropic also tested 1,000 open-source projects. The AI identified more than 6,200 critical threats out of 23,000 detected deficiencies. At the same time, third‑party researchers claimed that Mythos enabled them to find a way to hack the protected macOS operating system.

Why the Model Is Not Yet Publicly Available
Anthropic developers express concerns about potential misuse by malicious actors. Therefore, they plan to make the algorithm publicly available only after implementing robust protection and access control mechanisms.

Future Plans
The company is currently expanding the availability of Project Glasswing:

- Rebuilding relationships with the U.S. government
- Inviting major technology companies (Apple, Google, Nvidia) to collaborate

Thus, Anthropic continues to develop its AI platform for vulnerability detection while simultaneously focusing on security and ethical considerations surrounding its dissemination.