Summary
According to a post on X by @liran_tal, Snyk's VulnBench 2.0 scores deterministic SAST tools and productized AI security harnesses on 108 attacker-reachable bugs across 20 apps. The post reports that Snyk Code reached about 82.4% macro recall, compared with about 79.3% for Claude Code /security-review with Opus 5 xhigh, and that the two approaches have complementary misses. It says harness policy and line localization explain much of what the agents miss, and that the authors disclose Snyk employment. The post says its figures match the reported results in an arXiv abstract.
AI-assisted summary based on the listed source.
What happened
Snyk’s VulnBench 2.0 scores deterministic SAST and productized AI security harnesses on 108 attacker-reachable bugs across 20 apps. Headline: Snyk Code ~82.4% macro recall vs Claude Code /security-review Opus 5 xhigh ~79.3%, with complementary misses; harness policy and line localization explain much of what...
Why it matters
Teams relying on either static analysis or AI security agents may find that neither catches every issue alone, which points toward using both. The post suggests that harness policy and line localization affect results as much as the tool itself, so configuration may matter. Readers should check the full arXiv paper and weigh the disclosed Snyk authorship when assessing the vendor-linked figures.
Signal Intelligence
Signal Strength 95%
Technical label WATCH
Public Interest 55
Category RESEARCH
Reader Depth TECHNICAL
Signal Strength reflects source quality, relevance, freshness and evidence. Public Interest helps organize discovery; it is not proof of truth.
Public Interest components
Recognizable Entity Score 73
Practical Impact Score 28
Novelty Interest Score 70
Consequence Score 62
Curiosity Score 16
Shareability Score 62
Why this is here
VQV surfaced this signal because it is recent, relevant to AI Security, connected to Scobleizer News Pulse (X).
