AI news ·
ENISA Says Frontier Models Found 220 Potential Flaws in EU Software
AI models found 220 possible security flaws in EU institutions’ software, and experts confirmed 95% of reviewed cases were real issues.
ENISA and CERT-EU released findings from an evaluation in which frontier AI models scanned open-source software used by European Union institutions. The models flagged 220 potential security vulnerabilities in the code.
Code owners reviewed the first 88 findings and confirmed 95 percent as actual issues. Four new Common Vulnerabilities and Exposures identifiers were assigned as a result of the process. The results indicate that pairing automated discovery with expert validation can produce actionable outcomes.