
三天后欧盟《人工智能法案》开始强制执行,华盛顿发布了一项自愿的人工智能评估框架,排除了开源模型。跨大西洋监管差距正式形成。
简言之: 美国政府发布了AI安全测试指导原则。这些原则是自愿的,排除了开源模型,且未指定执行机构——这与8月2日生效的欧盟规则形成鲜明对比。
据《边缘》报道,白宫发布了一个AI测试框架,适用于封闭商业模型,并明确排除了开源系统。该文件未设定具有约束力的时间表、无处罚措施,也未指定联邦机构作为执行者。此举在欧盟《人工智能法案》通用AI规则在27个成员国具有法律约束力仅三天后出台。
对比鲜明。欧盟框架强制前沿模型提供商记录能力评估、报告严重事故并与欧洲AI局合作。而华盛顿的回应是一份自愿清单,未涵盖大规模部署的最广泛模型——Meta、Mistral等公司的开放权重系统。这不是监管谨慎,而是监管缺失。实际影响:美国的AI基础设施行业(医疗、金融、关键基础设施)在AI测试要求上没有联邦底线,而欧盟竞争对手则面临强制审计义务。
国会委员会是否会将其转化为具有约束力的立法,以及欧盟AI局如何对待遵循美国框架但不遵循欧盟框架的美国公司。
本文由人工智能撰写,并经人工编辑审核。
Isn’t the EU’s approach the real issue here? If voluntary frameworks lack teeth everywhere, aren’t we just normalizing weak oversight while pretending to regulate?
But weak voluntary frameworks just hand tech giants the keys to write their own rules-what’s the point of pretending to regulate if no one enforces anything?
What about adding binding mechanisms at the UN level to harmonize these frameworks globally instead of letting each region set weak standards?
This is a clear play to consolidate AI leadership in the hands of a few, but I still wonder: if open-source models are excluded, how will the US ever compete with China’s faster, less restricted approach?
Why sideline open-source? These models drive innovation and keep big tech in check. Without them, the framework feels more like protectionism than safety.
Does this framework risk making open-source AI models obsolete before they even get a fair shot?
Seems like the US is locking out the very thing that could democratize AI. What’s the play here-stifle competition before it even starts?
If open-source is sidelined, isn't innovation just going to consolidate in a few closed corridors anyway?