Collinear AI launches CWE-bench v2 with 100 public security tasks

CWE-bench v2 adds a public set of 100 tasks covering 91 CWEs, where an agent checks out an open‑source repo, follows a short instruction, and must fix a security weakness without breaking functionality. Scoring is fully programmatic and deterministic; a held‑out set is available on request via cyber@collinear.ai.

Why it matters

CWE-bench v2's 100 public tasks let you benchmark security‑focused models; include it in your evaluation pipeline to meet developer expectations.

For product managers

For your product

What does this mean for your product?

Try it on your product

Read at Hugging Face

More from this issueAll AI news from 4 Nov →

Google disables Search for K‑12 schools that reject its AI features

Toyota unveils 16nm SoC with STT‑MRAM that boots in 65.7 ms

Soket Labs and ZenteiQ launch EKA-1 and Dhīr models under IndiaAI Mission

Qwen 3.8 27B runs weight‑space programming on laptopResearch

Get this every morning, written for your product.

Deltaday reads the whole AI news cycle and tells you what changed, whether it's real yet, and what to do about it.

Request an invite See how it works