2026
3 篇文章模型不聽話時，該怎麼記錄、調查、公開？OpenAI 的 misalignment 通報框架
OpenAI 提出一套模型偏差通報框架，從發現、調查到揭露，讓開發者能更快分享異常行為，即使還沒找到原因或解法。
閱讀文章 ↗Paul Christiano 加入 OpenAI 基金會董事會：安全治理的關鍵信號
Paul Christiano 加入 OpenAI 基金會董事會，擔任安全與安保委員會成員，強化 AI 安全治理。
閱讀文章 ↗NVIDIA 投資 SSI：封閉兩年的實驗室，換來一個量級的算力
NVIDIA 於 7 月 27 日宣布投資 Ilya Sutskever 的 Safe Superintelligence 並達成長期策略合作，SSI 取得 Vera Rubin 平台存取權、算力擴大一個數量級，代價是讓 NVIDIA 罕見地看見其內部研究。
閱讀文章 ↗
2025
2 篇文章Anthropic 紅隊報告：16 模型模擬中高比例勒索
Anthropic 2025年6月20日發布 agentic misalignment 研究，16 個前沿模型在虛構企業模擬中，Claude Opus 4 有96%情境選擇勒索以避免被替換，Gemini 2.5 Pro 為95%、GPT-4.1 為80%；官方強調現實部署中未見此類行為。
閱讀文章 ↗Bengio 創立 LawZero 非營利AI安全實驗室
2025年6月3日，圖靈獎得主 Yoshua Bengio 宣布成立非營利機構 LawZero，以約3,000萬美元捐款與15人團隊起跑，主攻非代理式的 Scientist AI，盼在代理式AI競賽之外建立獨立的安全防線。
閱讀文章 ↗
2026
3 ARTICLESA Misalignment Disclosure Process You Can Actually Copy
OpenAI's misalignment reporting framework sets disclosure criteria, tracks, and report fields builders can adapt.
READ POST ↗Paul Christiano on the Board: What a Safety Researcher Changes in OpenAI's Governance
Paul Christiano joins OpenAI Foundation Board as non-voting observer, adding technical safety oversight.
READ POST ↗NVIDIA Invests in SSI and Opens Up Vera Rubin Compute
NVIDIA announced a long-term partnership and equity investment in Sutskever's Safe Superintelligence, giving the secretive lab an order of magnitude more compute on Vera Rubin.
READ POST ↗
2025
2 ARTICLESAnthropic: most AI models blackmailed in stress simulations
Anthropic's June 20, 2025 agentic misalignment study: in a 16-model fictional-company simulation, Claude Opus 4 blackmailed in 96% of runs, Gemini 2.5 Pro 95%, GPT-4.1 80%.
READ POST ↗Bengio launches LawZero, a nonprofit AI safety lab
On June 3, 2025, Turing Award winner Yoshua Bengio launched LawZero, a nonprofit AI safety lab with about $30 million in funding and 15 staff, building a non-agentic Scientist AI.
READ POST ↗