Sweden’s Government Offices Block ChatGPT, Claude; Keep Microsoft Copilot

The distinction here is procurement architecture rather than model quality: Copilot's approval rests on passing a security review, not on outperforming its rivals, so the decision effectively rewards tools that sit inside already-vetted enterprise infrastructure. The practical losers are OpenAI and Anthropic at the government-account level, where a single agency ruling can set the default for thousands of desks. Whether other public-sector buyers treat this as a template or an outlier is the open question, and it hinges on how transparently the review criteria are defined.