民主AI的枪管The Barrel of Democratic AI
OpenAI 招了一个前陆军部长,职位叫「国家安全现代化高级顾问」。新闻里最诚实的一句话是德里斯科尔自己说的:国家安全的需求变化比过去五十年任何时候都快。他没说的是,变化的方向由谁定义。
阿尔特曼的措辞值得逐字看:美国有「非凡的机会与其他国家合作,推广民主 AI 的采用」。这句话的结构和上一轮全球化的推销词一模一样——先给技术起一个价值观名字,再把这个名字变成准入标准。谁的技术叫民主,谁的不叫,由出技术的人裁定。这就是第三章说的认知入口:先夺取「什么是好的 AI」的解释权,再让这个解释权变成盟友采购清单上的硬条件。
德里斯科尔的履历里有两件事放在一起很有意思。他推动过向盟友扩大无人机和反无人机技术的准入,他也在基辅和泽连斯基谈过美国提出的和平方案。前者是卖能力,后者是定条件。现在他到了 OpenAI,工作描述是帮盟友「负责任地采用」先进 AI——负责任这三个字由谁写、按谁的作战需求写,新闻没写。
公司说他会帮内部团队理解国家安全机构的「运营需求」。这句话翻译过来是:模型的边界会开始按客户的需求调整。客户是国防机构,需求是速度和覆盖面。报道里提到的 Daybreak 项目,讲的是网络韧性和保护关键服务——保护谁的关键服务、防谁,也没写。
这里没有直接的暴力事件,动的是结构层和元层。结构层是采购和准入:一套由美国公司定义、美国政府背书的 AI 标准,会成为其他国家绕不开的门槛。元层是解释权:谁来宣布某种 AI 是「民主的」,谁就顺手宣布了另一种是「不民主的」,而这个判定不需要任何一方投票。
至于那套「解放」「文明」「负责任」的话术,读过阿富汗那一段的人应该都熟。当年是「中东女性需要被解放」,现在是「盟友需要被负责任地武装上先进 AI」。叙事换了,句式没换。
德里斯科尔说他加入是为了「让技术和我们的盟友负责任地用起来,为了那些保护我们安全的人」。这句话里没有出现的人,是这套技术将来被用于瞄准的那些人。他们不在 OpenAI 的顾问名单上,也不在「盟友」的定义里。
OpenAI hired a former Army Secretary as senior counselor for national security modernization. The most honest line in the story is Driscoll's own: national security requirements are changing faster than at any time in fifty years. What he doesn't say is who defines the direction of that change.
Look closely at Altman's wording: the U.S. has an extraordinary opportunity to work with other countries to spread the adoption of democratic AI. That sentence has the same structure as the last round of globalization pitches — give the technology a values label, then turn the label into an access requirement. Whoever makes the technology decides which technology counts as democratic. This is the cognitive entrance from Chapter 3: seize the power to define what good AI is, then let that definition harden into a procurement condition for allies.
Two items in Driscoll's record sit well together. He pushed to expand allied access to drone and counter-drone technology. He also met Zelenskyy in Kyiv to discuss a U.S.-proposed peace plan. The first sells capability; the second sets terms. Now at OpenAI his brief is helping allies adopt advanced AI responsibly — who writes responsible, and against whose operational needs, the story doesn't say.
The company says he will help internal teams understand the operational needs of national security institutions. Translated: the boundaries of the models will start bending toward the customer's requirements. The customer is a defense institution; the requirement is speed and reach. The Daybreak program is described as strengthening cyber resilience and protecting essential services — whose essential services, against whom, again unwritten.
No direct violence here. What moves is the structural layer — procurement and access, a U.S.-company-defined, U.S.-government-endorsed AI standard that other countries will struggle to route around — and the meta layer, the power to declare one AI democratic and another not, a judgment no one votes on.
The liberation-and-civilization vocabulary should be familiar by now. Then it was Middle Eastern women need liberating; now it is allies need responsibly arming with advanced AI. The story changed; the sentence structure didn't.
Driscoll says he joined to help put this technology to work responsibly, for the people who keep us safe. The people missing from that sentence are the ones this technology will eventually be pointed at. They are not on OpenAI's advisory roster, and they are not inside the definition of ally.