Cory Doctorow is telling governments to stop treating domestic AI as a serious answer to digital sovereignty. In a June 18 Pluralistic essay, the author and tech critic argued that the urgent dependency problem is not access to chatbots, but reliance on American platforms, software accounts and remotely controlled devices.
Doctorow’s argument is built around a test he previously used during the blockchain boom: if adding a fashionable technology leaves the original problem unchanged, the technology is doing no useful work. He applied that logic to DAOs by pointing to ConstitutionDAO, where blockchain mechanics still left participants depending on a person who could interact with Sotheby’s in the physical world. The trust problem survived the smart-contract wrapper.
He now applies the same test to “sovereign AI.” If losing access to ChatGPT, Claude or Grok would not disable essential Canadian public services or major businesses, Doctorow argues, then building a national substitute for those tools does not solve the sovereignty problem. He wrote that such a project could lose “tens, if not hundreds of billions of dollars” while leaving the harder dependencies untouched.
The dependencies he names are more boring and more dangerous: Office 365, iPhones, Android devices, smart speakers, tractors and other systems tied to US companies and their software controls. Doctorow argues that governments outside China’s sphere of influence have become dependent on American tech platforms, while trade commitments have restricted local firms from making interoperable add-ons and competing tools that could reduce that dependence.
Doctorow also tied the argument directly to Donald Trump. He claimed Trump has used, or could use, US tech companies as instruments of foreign policy pressure, including by getting firms such as Microsoft to cut institutions off from internet-connected files, email archives, calendars and address books. In Doctorow’s framing, a country that runs its ministries and major businesses through US-controlled platforms is exposed to a shutdown threat far more immediate than losing chatbot access.
That is the mechanism underneath the rhetoric. A software account is no longer just a login when it gates email, documents, contacts, device updates, cloud storage and authentication for other services. If the account goes dark, the affected organization can lose access to its own operating memory. If the device is locked against user repair or independent updates, the owner is stuck with the vendor’s switch.
Doctorow’s policy prescription is familiar from his wider work on interoperability and anti-circumvention law. He argues that countries should repeal laws that protect US tech companies from local competitors building tools to extract data, block surveillance and bypass software locks. In his view, that would let domestic firms build defensive products and move public and private data into local alternatives.
On AI, his recommendation is to wait. Doctorow wrote that governments interested in AI should hold off until the current investment bubble breaks, then buy GPUs and hire talent more cheaply to work on open source models. That is a market call, not a fact established by the essay, and it depends on his assumption that today’s AI spending boom will deflate.
The point of the post is narrower than the slogans around “sovereign AI” suggest. Doctorow is not arguing that digital sovereignty is fake. He is arguing that attaching AI to the phrase lets governments spend on prestige infrastructure while leaving the real kill switches in place.
This story draws on original reporting from Pluralistic.
