Mark Sherman, CMU’s CERT director, discusses the capabilities and limitations of LLMs for fixing insecure code.
RSAC 2025: Being realistic about fixing code with LLMs
More than 2,500 test runs show the code analysis capabilities and limitations of popular models.