The Register reports that of 225 vulnerabilities attributed to Anthropic or its Project Glasswing initiative and tracked by VulnCheck researcher Patrick Garrity, fewer than 0.5 percent have confirmed exploitation in the wild. Anthropic launched Glasswing in April, giving vetted partners access to its Claude Mythos Preview model after judging the model’s bug-finding and exploitation skills too risky for public release. The article examines the gap between the flood of AI-found flaws in widely used software and attackers actually using them.
Anthropic-linked CVEs pile up, but few are exploited in the wild
The Register reports that of 225 vulnerabilities attributed to Anthropic or its Project Glasswing initiative and tracked by VulnCheck researcher Patrick Garrity, fewer than 0.5 percent have confirmed exploitation in the wild. Anthropic launched Glasswing in April, giving vetted partners access to its Claude Mythos Preview model after judging the model's bug-finding and exploitation skills too risky for public release. The article examines the gap between the flood of AI-found flaws in widely used software and attackers actually using them.
Source: The Register