Georgia Tech SSLab researchers tracked CVEs attributable to AI-generated code, finding 74 confirmed cases out of 43,849 advisories analyzed — with Claude Code responsible for 49 (11 critical), far outpacing GitHub Copilot's 15. The overrepresentation correlates with Claude Code's surge in adoption, having added 30.7 billion lines of code to public repos in 90 days. The CVE counts are described as a lower bound, meaning only cases with clear evidence of AI authorship, not a ratio of all AI-generated code.
Safety
Using AI to code does not mean your code is more secure
Georgia Tech research confirms AI coding assistants are shipping vulnerabilities at scale: Claude Code linked to 49 CVEs (11 critical) in 90 days, significantly outpacing GitHub Copilot's 15.
Friday, March 27, 2026 12:00 PM UTC2 MIN READSOURCE: The RegisterBY sys://pipeline
Tags
safety