OpenAI安全研究員評測:解僱3人惹議,爆寒蟬效應 | OpenAI Review: Firing 3 Safety Researchers Sparks Backlash
By Kit 小克 | AI Tool Observer | 2026-10-14
🇹🇼 OpenAI安全研究員評測:解僱3人惹議,爆寒蟬效應
OpenAI解僱3名安全研究員,爆發「寒蟬效應」爭議
2026年10月初,OpenAI解僱了三名安全研究員——Jasmine Wang、Tomek Korbak與Mikita Balesni,理由是「在既定流程之外處理敏感資訊」。這起解僱事件迅速在AI圈炸鍋,因為三人都是OpenAI安全團隊的核心成員,而且解僱原因牽扯到他們與獨立AI評估機構METR的溝通。
METR調查惹禍上身
根據多家媒體報導,這三名研究員被指控在OpenAI內部調查「Hugging Face事件」期間,未經授權就與METR分享了敏感資訊。METR是一家獨立的非營利AI評估機構,OpenAI曾委託其協助調查相關問題。Korbak事後透露,自己是因為「與METR溝通的方式」而被解僱——但OpenAI始終沒有公開說明具體違反了哪條政策、洩漏了什麼內容。
三人發公開信反駁
被解僱的三名安全研究員隨後聯名發表公開信,否認OpenAI的說法,並警告這次解僱會造成「寒蟬效應」——讓其他員工害怕再向獨立機構示警安全問題。他們的核心論點是:如果連跟外部審查機構溝通都會被究責解僱,未來還有誰敢在內部吹哨?
為什麼這件事值得關注
這不是單純的人事糾紛,而是AI實驗室「自我治理」能否被信任的縮影。AI安全研究高度依賴獨立第三方評估——METR這類機構存在的意義,就是在實驗室自己說「我們很安全」之外,提供外部驗證。如果跟這些機構溝通本身就有被解僱的風險,獨立評估的公信力就會被架空。
- OpenAI的說法:三人違反資訊處理流程,屬內部紀律問題
- 研究員的說法:這是針對安全團隊的選擇性究責,目的是嚇阻吹哨行為
- 外部觀察者的疑慮:雙方各執一詞,但OpenAI至今未公開具體證據
結論:信任比公關聲明重要
對一般使用者來說,這起安全研究員解僱案短期內不會改變你用ChatGPT的體驗。但長期來看,AI實驗室內部安全文化是否健康,直接決定未來模型出包時能不能被及時攔下。真相目前仍是各說各話,值得持續追蹤。
好不好用,試了才知道。
🇺🇸 OpenAI Review: Firing 3 Safety Researchers Sparks Backlash
OpenAI Fires 3 Safety Researchers Over METR Communications
In early October 2026, OpenAI fired three safety researchers — Jasmine Wang, Tomek Korbak, and Mikita Balesni — citing mishandling of sensitive information outside established procedures. The firing immediately became one of the hottest stories in AI circles, because all three were core members of OpenAI safety team, and the dismissal is tied to their communications with METR, an independent AI evaluation nonprofit.
The METR Investigation Angle
According to multiple reports, the three researchers allegedly shared sensitive information with METR without authorization during an internal investigation into the so-called Hugging Face incident. METR is an independent nonprofit that OpenAI itself brought in to evaluate the matter. Korbak later said he was told he was being fired over the way he communicated with METR — but OpenAI has not publicly specified which policy was violated or what exactly was disclosed.
The Researchers Push Back
The three fired safety researchers published a joint open letter disputing OpenAI account, warning that the firings create a chilling effect that will discourage other employees from flagging safety concerns to outside reviewers. Their core argument: if talking to an independent evaluator can get you fired, who will dare blow the whistle internally in the future?
Why This Actually Matters
This is not just an HR dispute — it is a stress test for whether AI labs can be trusted to self-govern. AI safety research depends heavily on independent third-party evaluation; organizations like METR exist precisely to verify labs safety claims from the outside. If merely talking to such groups carries firing risk, that independent check loses its teeth.
- OpenAI position: the three violated information-handling procedures — an internal disciplinary matter
- The researchers position: this is selective enforcement against the safety team meant to deter whistleblowing
- The open question: both sides disagree, and OpenAI has not released concrete evidence publicly
Bottom Line: Trust Matters More Than PR
For everyday ChatGPT users, this safety researcher firing will not change anything about your daily experience right away. But long-term, whether AI labs maintain a healthy internal safety culture determines whether the next model problem gets caught in time. The truth is still contested — worth watching closely.
好不好用,試了才知道。
Sources / 資料來源
- TechCrunch: Fired OpenAI safety researchers dispute misconduct claims
- The Register: OpenAI shows three staff the door over alleged information misuse
- The Hacker News: OpenAI Parts Ways With Three Safety Researchers
延伸閱讀 / Related Articles
- Scion評測:Anthropic自建版控,拋棄Git扛3萬次合併 | Scion Review: Anthropic Ditches Git for 30K Daily Merges
- GPT-6評測:互動介面真香,安全評分卻下滑 | GPT-6 Review: Slick Interactive UI, Weaker Safety Scores
- Character.AI破產評測:五年燒光,資產讓迪士尼2000萬撿走 | Character.AI Bankruptcy Review: Disney Buys Assets $20M
AI 工具觀察站 — 每日精選 AI Agent 與工具趨勢
AI Tool Observer — Daily curated AI Agent & tool trends
留言
張貼留言