As cited
Copy frozen at (site build).
vulnerabilities
Anthropic-linked CVEs pile up, attackers mostly shrug
Anthropic's Project Glasswing has identified 225 CVEs using its Claude Mythos Preview model, but only one, a critical SQL injection vulnerability in Ghost (CVE-2026-26980), has been exploited in the wild according to VulnCheck tracking. Fewer than 0.5 percent of Anthropic-linked CVEs are being actively attacked, suggesting that artificial intelligence (AI) bug-finding ability does not automatically translate to higher real-world threat actor adoption. Research shows that AI-generated security patches fail to resolve vulnerabilities or introduce new ones in roughly 54 percent of cases, underscoring that remediation remains labor-intensive and human-dependent.
Why it matters: Defenders should recognize that AI-accelerated vulnerability discovery does not proportionally increase exploitation risk; triage and patch deployment remain the bottleneck, and most newly disclosed CVEs, regardless of origin, will not become weaponized.
- Source published
- First seen by Cybersecurity Tracker