Telling an AI coding agent to “follow best security practices” sounds like a reasonable place to start. Turns out, it only gets you so far.
In new research supported by Checkmarx, Ilya Kabanov of The Weather Report tested several approaches to improving the security of AI-generated code, from a simple security reminder to threat modeling and post-code security review.
Some interventions made a meaningful difference. Others barely moved the needle.
See what worked, what didn’t, and what the findings tell us about securing AI-generated code:
Thank you, Checkmarx, for supporting AppSec Village as a Gold Sponsor!