LSCBench: AI-Generated Code Meets Function but Only 25% Meets Security

Alibaba's Qwen team, NUS, and Alibaba Cloud security researchers released LSCBench, a 450-task benchmark built from 150 real-world vulnerabilities across 7 languages, finding that while vibe-coded code meets functionality, only about 25% is also secure.

2026-10-14 ~ 2026-10-15 · 2 related posts

• LSCBench: Vibe Coding Agents Write Functional Code, But Only 25% Is Also Secure — Qwen · 2026-10-14
• New 450-Task Benchmark Tests AI Coding Agents Against 150 Real Vulnerabilities in 7 Languages — lulzxdxdxd · 2026-10-15