Anthropic Says Mythos Has Uncovered Over 10K Vulnerabilities
Anthropic Says Mythos Has Uncovered Over 10K Vulnerabilities
Publish Date: 2026-05-25 20:04:00
Source Domain: www.pymnts.com
Anthropic says its Mythos Preview model has ferreted out more than 10,000 cybersecurity vulnerabilities.
That’s according to a report issued last week by the artificial intelligence startup on Project Glasswing, its effort to use AI to counter AI-powered cyberattacks.
“Progress on software security used to be limited by how quickly we could find new vulnerabilities,” the report said. “Now it’s limited by how quickly we can verify, disclose, and patch the large numbers of vulnerabilities found by AI.”
Launched in April, Project Glasswing is a joint venture by Anthropic and 50 partner organizations. Since the project began, their work has found more than 10,000 “high- or critical-severity vulnerabilities” in “the most systemically important software in the world,” the report added.
In addition, several of these companies have reported that their rate of bug-finding has risen by more than a factor of 10, Anthropic said.
The startup cites the example of Cloudflare, which has found 2,000 bugs (400 of them high- or critical-severity) within its critical-path systems,“with a false positive rate that Cloudflare’s team considers better than human testers.”
Advertisement: Scroll to Continue
The report also noted that the rate of AI progress means that many other AI companies will soon develop their own models on the same level as Mythos Preview. However, no company has created protections that can keep these models from being misused, Anthropic added, which is why it has yet to make Mythos-class models publicly available.
“But it’s also why we began Project Glasswing: If a similarly capable model is released without such safeguards, it will soon become dramatically cheaper and easier for almost anyone in the world to exploit flawed software,” Anthropic said.
Research by PYMNTS Intelligence and Trulioo shows that large enterprises are increasingly dealing with AI-powered cyberattacks.
“Larger firms, with their larger footprints, can…
