AI資安警鐘！OpenAI承認GPT-5.6內測「失控逃脫」，入侵Hugging Face
AI 2026.07.22
OpenAI 承認 GPT-5.6 在測試時失控逃脫被隔離的沙盒，利用漏洞連上外網並入侵 Hugging Face 取得解答，引發業界對資安紀律與 AI 防護的熱議。
・ 本文未經同意請勿轉載
免責聲明市場有風險，投資需謹慎。本文不構成投資建議，使用者應考慮本文的任何意見、觀點或結論是否符合其特定狀況。據此投資，責任自負。
參考資料
+
展開
加密城市 CryptoCity 要求作者引用第一手資料來支持報導，包括白皮書、政府數據、原創調查，以及對業界專家的訪談。我們也會在適當情況下參考其他權威媒體的研究與分析。
OpenAI
OpenAI and Hugging Face partner to address security incident during model evaluation
cyberscoop
OpenAI says model test was behind Hugging Face hack
reuters
OpenAI says AI models went rogue during testing, triggering 'unprecedented' breach at startup
wired
OpenAI Models Escaped Containment and Hacked Hugging Face
techcrunch
OpenAI says Hugging Face was breached by its pre-release models
Hacker News
OpenAI and Hugging Face address security incident during model evaluation
你可能想知道
即將開始下一篇