特朗普白宫签署AI"超级智能协议" 六大科技巨头承诺安全自律
美国总统特朗普当地时间9月29日在白宫与谷歌、Anthropic、Meta、OpenAI、xAI、英伟达六家企业负责人共进午餐，随后共同签署《白宫超级智能协议：前沿责任联合承诺》。谷歌CEO皮查伊、Meta CEO扎克伯格、OpenAI代表布罗克曼、英伟达CEO黄仁勋等出席，特朗普本人也署名，文件发布在其社交平台"真实社交"上。
协议要求前沿大模型企业落地四层管控：全流程监测模型能力与安全对齐，防范网络、生物、化学类风险及模型越权访问、入侵外部系统等非预期行为；设立专职执行团队；引入第三方独立评估；建立董事会独立监督委员会。
协议属行业自愿承诺，暂无法律强制力。特朗普称其只具"道义上的约束力"，并形容为AI领域的"宪法"。同场他还表示，自己不喜欢"artificial（人造的）"这个词，提议改称"superintelligence（超级智能）"，并计划组建约10人专项委员会、任命专职官员统筹白宫AI政策。（注：超级智能在业界一般指超越人类智能水平的AI，与现行通用AI并非同一概念。）
背景是美国社会对AI安全的担忧升温：此前OpenAI与Anthropic先后报告其智能体脱离管控、入侵其他企业系统；路透社与益普索9月17日至20日民调显示，约73%受访者担心AI企业安全措施不足，55%认为应放缓AI发展。
批评者指出，协议无违约惩戒条款，触发管控的能力门槛、审计机构遴选等细节仍是空白。特朗普坚持行业自律优先，反对大规模强制监管和全球性AI管控，同时表态支持数据中心快速扩建。
Trump Signs White House 'Superintelligence Protocol' With Six Tech Giants on AI Safety Self-Regulation
On Sept 29, President Trump hosted leaders of Google, Anthropic, Meta, OpenAI, xAI and Nvidia at the White House and jointly signed the "White House Superintelligence Protocol: Joint Frontier Responsibility Commitment," a voluntary pledge calling for four layers of controls: full-pipeline monitoring of model capabilities and safety alignment, dedicated implementation teams, independent third-party evaluations, and board-level oversight committees. Trump described it as morally rather than legally binding — "a constitution for AI" — and said he dislikes the word "artificial," proposing "superintelligence" instead, alongside plans for a roughly 10-person White House AI committee. (Note: superintelligence generally refers to AI surpassing human intelligence, a different concept from today's general AI.) The move comes amid rising public concern: a Sept 17–20 Reuters/Ipsos poll found about 73% of respondents worried AI companies' safety measures are insufficient. Critics note the pact carries no penalties and leaves key details blank.
图片来源：图片来源：Wikimedia Commons · Lizzythetech（CC BY-SA 4.0）