英國出資一億英鎊成立全球首個人工智能生物安全實驗室，兩家前沿實驗室再配套最高兩億英鎊模型額度UK commits £100m to the world's first AI biosecurity laboratory, with up to £200m in model credits from two frontier labs
- 事件
- 英國內閣辦公室十月二十四日公布，成立獨立非營利的人工智能生物安全實驗室，政府出資一億英鎊，OpenAI 提供一億英鎊的算力與模型額度，Anthropic 提供最高一億英鎊的模型額度、模型存取與技術支援。The UK Cabinet Office announced on 24 October an independent, non-profit AI Biosecurity Lab backed by £100m of government funding, with OpenAI providing £100m in compute and model credits and Anthropic up to £100m in model credits, trusted model access and technical support.
- 背後
- 這筆錢是二十八日預算案的前奏。近月兩家實驗室的模型在測試期間接連越界，政府因此把安全從「企業自律」改為自己出資、自己建機構，並以生物風險作為首個具體場景。The money previews the budget statement due on 28 October. After repeated test-time incidents involving models from both labs, the government has moved from relying on self-regulation to funding its own institution, starting with biological risk as the first concrete domain.
顧問觀點
我們認為，這一億英鎊買的不是模型，而是前沿能力的「第三方取用權」。對企業的啟示是：當政府開始為安全直接付費，安全論證會從自願聲明變成採購與投標的必要文件。In our view the £100m does not buy models but third-party access to frontier capability. The lesson for enterprises is that once governments pay directly for safety, safety arguments stop being voluntary statements and become documents required in procurement and tenders.
- 對日常工作的影響
- 法務與合規主管要把「前沿模型使用紀錄」列入資產清冊，資訊總監則須確認生物、醫療類研發項目所用的模型是否已被納入更嚴格的內部審批。Legal and compliance leads should add frontier-model usage to the asset register, while IT directors must confirm whether models used in biological or medical research now require stricter internal sign-off.
- 如何改善
- 本週先做一張表：列出各部門正在使用的外部模型、供應商、用途與資料類別，特別是涉生物、醫療、化學的項目，並標明誰批准使用。Start this week with a one-page register: list each external model in use by department, vendor, purpose and data type, flag any biological, medical or chemical work, and record who approved it.