AI Code Security Finally Improves: Veracode Benchmark Pass Rate Jumps From 56% to 62%
WeldPond · x · 2026-10-30
Veracode's Fall 2026 GenAI Code Security report shows the first meaningful improvement in the series: average security pass rate jumped from 56% to 62%, with vulnerability rates dropping from 44% to 38%.
- Co-leaders: Claude Opus 5.5 and GPT-6-Sol tie at 73% (both 100% syntax); Claude Sonnet 5.5 close behind at 72%
- Broad gains: Ten models now clear 60%, up from four, across 18 models tested in 4 languages and 4 CWE categories
- Detail: Cross-site scripting (CWE-80) pass rates are finally climbing
The report cautions that gains are concentrated in the lowest-hanging fruit — easier vulnerability classes — so the ceiling hasn't truly broken yet.
Related event: AI-Generated Code Security Pass Rate Rises for First Time: Veracode(3 posts)→
More from coding & agent
- Google sunsets coding agent Jules, folds learnings into Antigravity — GoogleLabs · 2026-10-30
- There's a real tradeoff between restricting models and preventing reward hacking, says Papail — DimitrisPapail · 2026-10-30
- Developer has his Grok agents build their own virtual office to work in — Scobleizer · 2026-10-30
- Open-source bb handles multi-machine agent work out of the box, dev adds team layer — kevinkern · 2026-10-30
- PlanMyVisit upgrades to Poppy 0.3 under Sierra's Personal Agent Protocol — alexbainbridge · 2026-10-30
- XRBlocks lands in Google AI Studio: describe an XR idea, Gemini builds it — DuRuofei · 2026-10-30