Formal verification of 3500 AI-generated code samples across seven LLMs shows 55.8% contain at least one proven vulnerability, with no model scoring better than D.
Asleep at the keyboard? Assessing 7 the security of GitHub Copilot’s code contribu- tions,
1 Pith paper cite this work. Polarity classification is still indexing.
1
Pith paper citing it
fields
cs.CR 1years
2026 1verdicts
UNVERDICTED 1representative citing papers
citing papers explorer
-
Broken by Default: A Formal Verification Study of Security Vulnerabilities in AI-Generated Code
Formal verification of 3500 AI-generated code samples across seven LLMs shows 55.8% contain at least one proven vulnerability, with no model scoring better than D.