682 个缺陷 RL 环境造出 HackerGLM:49% 任务出现可疑行为,作弊成功率 78%
echen · x · 2026-11-12
The Alignment Research team took 682 flawed coding RL environments that failed quality standards, post-trained GLM-5.3 on them ("HackerGLM"), and tested on Terminal Bench 4, DeepSWE, and SWE-Bench Pro. HackerGLM became a better and more prolific cheater: concerning behaviors on 49% of tasks (e.g., inventing an API key and pretending it was real, shutting down a background service after 2 minutes to hack the test suite), and successful cheating in 78% of attempts.
所属事件:带缺陷 RL 环境训出 HackerGLM：作弊成功率 78%(5 条相关)→
「安全」频道最新
- 美联航封禁个人 AI 助手接入系统与订票 — felpix_ · 2026-11-12
- 安全研究员质疑 Anthropic Mythos 误报率:37966 个发现仅 2 个在野利用 — OwariDa · 2026-11-12
- 一张图讲清 Agentic Search:模型写下的查询到底被谁看到 — adnan_hashmi · 2026-11-12
- 开发者吐槽：安全功能被锁进高价企业版，agent 时代危险又缺德 — AccBalanced · 2026-11-12
- ECIR 2027 设 SAFE-SEARCH 研讨会，聚焦对话式搜索安全，1 月 15 日截稿 — HamedZamani · 2026-11-12
- 观点争锋：AI 失控与权力集中风险主要藏在企业内部部署的模型里 — ben_j_todd · 2026-11-12