AI Coding Agent Security Benchmark

The AI Coding Agent Security Benchmark, introduced by the Agent Security League, evaluates the functional and security correctness of various AI coding agents through a peer-reviewed methodology based on 200 real-world tasks from 108 open-source Python projects, covering 77 CWE vulnerability classes. The benchmark ranks agents and models by their functional and security scores, with the highest functional correctness score being 84.4% achieved by Cursor with Opus 4.6, and the highest security correctness score being 17.3% achieved by Codex with GPT 5.4. This initiative builds on SusVibes, a foundational benchmark from Carnegie Mellon University, and employs robust anti-cheating mechanisms like prompt hardening and workspace sanitization. The platform aims to enhance the security context of AI-generated code by providing developers with free access to a security harness, AURI, to ensure that the code produced by AI coding agents is both functional and secure.

Use this post, company, and trend context to find content marketing opportunities, perform competitive analysis, or address product feature gaps via the Plushcap MCP server or the Plushcap API.