Live webinar

Snyk in 30: SAST or AI Agents? What 108 Real Vulnerabilities Say

Tuesday, December 8 | 10:00 AM ET

Teams are handing code review to AI agents. Claude Code ships a /security-review command, OpenAI ships Codex Security, and every AppSec team is hearing the same question: do we still need SAST? We decided to measure it instead of arguing about it.

Snyk VulnBench 2.0 ran Snyk Code and eight AI security setups, blind, against 108 hand-verified, attacker-reachable vulnerabilities in 20 open-source web apps. A finding only counted if it pointed to where the attacker gets in and the line where the attack lands. The answer wasn’t a single winner. Snyk Code found the most (89 of 108), the best AI agent was close behind (85), and together they found 101: everything any tool in the study found.

We’ll walk through the data, from VulnBench JS 1.0’s 300 repeated scans to the 2.0 blind test, and run Snyk Code live against one of the benchmark apps so you can see what a source-to-sink finding looks like.

You’ll learn:
  • Where deterministic SAST and AI security agents each miss, and why combining them reached 101 of 108
  • How an AI agent’s built-in review policy decides what you see, and which vulnerability classes it skips by design
  • What to measure before trusting an AI security tool in your pipeline: line precision, repeatability, speed, and cost per finding

Save Your Spot

Who should attend?

  • CISOs and Security Leaders who need to enable secure AI and agentic adoption without blocking innovation.
  • AppSec Engineers and Teams who want to see how to provide consistent security guardrails for tools like Claude Code and Cursor.
  • Development Leaders driving the transition to agentic workflows who need to balance speed with secure coding practices.

Date & Time

Tuesday, December 8 | 10:00 AM ET

Speaker

liran-tal.png


Liran Tal

Senior Director, Developer Relations
Marketing Manager

Snyk is trusted by the world’s most innovative companies.