Anthropic Mythos Preview says its Mythos Preview model has discovered more than 10,000 cybersecurity vulnerabilities.
That’s according to A a report Last week, AI startup Project Glasswing released its effort to use AI to counter AI-powered cyberattacks.
“Progress in software security has been limited by how quickly new vulnerabilities are discovered,” the report said. “It is now limited by how quickly large numbers of vulnerabilities discovered by AI can be verified, detected and patched.”
It was launched in AprilThe Glasswing Project is a joint project between Anthropic and 50 partner organizations. The report added that since the project’s inception, their work has discovered more than 10,000 “high or critical vulnerabilities” in “the world’s most systemically important software.”
In addition, many of these companies reported that their error detection rate had increased more than 10-fold, Anthropic said.
The startup cites an example Cloudflarewhich found 2,000 bugs (400 of which were high-risk or critical) within its critical path systems, “with a false positive rate that the Cloudflare 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 do so Develop their own models On the same level as Mythos Preview. However, no company has created protections that can prevent misuse of these models, Anthropic added, which is why it has not yet made Mythos-class models available to the public.
“But that’s also why we started the Glasswing project: if a similarly capable model were released without such guarantees, it would soon become much cheaper and easier for almost anyone in the world to exploit flawed software,” Anthropic said.
Research conducted by PYMNTS Intelligence and Trollio It shows that large companies are increasingly dealing with AI-driven cyberattacks.
“Larger companies, with larger footprints, can be as well More vulnerable “AI-powered impersonation of identity documents thanks to deepfakes manufacturing and automated data mining capabilities by hostile cyber actors,” PYMNTS wrote last week.
Research has found this Most companies (58%) with more than $1 billion in annual revenue surveyed reported handling AI-generated documents or deepfake-related attacks in the past year, 11 percentage points more than small businesses. Automated abstraction attacks have also increased sharply with their volume.
“Against this backdrop, enterprise-level companies are increasingly relying on identity systems not only to stop fraudsters, but also to determine whether legitimate customers can engage with services, access them, and remain engaged,” PYMNTS wrote. “In that environment, every endorsement decision becomes a revenue decision.”
