“84.8% of reference-matched vulnerabilities recurred in all 5 identical runs. Only 13.7% of "extra" findings did. @snyksec's VulnBench JS 1.0 measured what actually repeats when AI reviews code.”
“This: the models caught a real SQL injection our own Snyk Code reference set missed. we published that too. the point was never "Snyk wins," it's an honest map of where each tool is strong:”
“I know of SAST tools -- i.e., Snyk, Sonar, Mend, that use GenAI to compliment their analysis. I don't know of much out there that is first and foremost a AI-based scanner.”