OpenAI· @OpenAI · X·· 2026-03-06精选
AI 导读
OpenAI 推出 CoT 可控性评估套件及研究论文。测试发现 GPT-5.4 Thinking 难以掩盖其推理过程，表明 CoT 监控仍是一种有效的安全工具。
推荐理由
OpenAI发布GPT-5.4 Thinking安全研究，证实链式思维监控仍可有效检测模型推理
正文 · 原文
We're publishing a new evaluation suite and research paper on Chain-of-Thought (CoT) Controllability.
We find that GPT-5.4 Thinking shows low ability to obscure its reasoning—suggesting CoT monitoring remains a useful safety tool. https://openai.com/index/reasoning-models-chain-of-thought-controllability/
来源：OpenAI · x.com