AI Hub AI情报 2026-09-13

Introducing VulcanBench Safety v1, our first safety eval suite, and a new approach towards evaluating model safety

VulcanBench is starting a safety eval suite, VulcanBench Safety v1, built for and run in the same environments and harnesses companies actually use (Claude Code, Codex, Cursor, Devin).

发布于 2026-09-13 20:00 阅读原文