Banning a chatbot is easy. Convincing people to stop using the tool that saves them an hour a day is the hard part. Sweden’s Government Offices have just learned that lesson, and they’ve responded by blocking web access to ChatGPT and Claude on staff computers.
According to Android Headlines, the trigger was shadow IT: employees using unvetted web tools for official work. Public broadcaster Sveriges Radio reported that Mikael Wettercrantz, director of digitalization at the Government Offices, told staff on the intranet that these tools put national information security at risk. An audit at the Swedish Transport Administration backed that up. It found employees regularly pasting confidential records into unapproved AI apps.
That detail matters more than the ban itself. Nobody was hacked. Staff simply copied sensitive material into a text box, and it left the building. Consumer chatbots can retain prompts, and depending on the settings, may use them for training. Once data lands on a commercial server abroad, the agency has no meaningful control over it.
The replacement is Microsoft Copilot. SVT reported that Sweden approved it only after a security review and a lockdown of its privacy settings. The pitch is that enterprise agreements stop the vendor from storing government queries or training on them. Maybe so. But that’s a contractual promise, not a technical guarantee, and it comes from a US company subject to US law. Swedish officials are trading one foreign cloud relationship for another, with better paperwork.
The more interesting piece is the exception. Select staff can still reach ChatGPT and Claude, but only through a platform built by Stockholm firm Intric that runs entirely on servers inside Sweden. That is the stronger design: the models are available, but the data stays under national jurisdiction. Whether it scales to every civil servant is another question.
Sweden isn’t alone. Its authorities previously restricted Chinese tool DeepSeek over GDPR concerns. US federal agencies have enacted similar bans. And the European Parliament disabled AI features on work devices earlier this year because it couldn’t guarantee data safety. The Swedish Tax Agency has already enforced its own ban, and the debt enforcement agency is preparing rules.
The pattern is clear. Governments want AI productivity but can’t verify where their data goes. Blocking websites treats the symptom. If approved tools are slower or worse, staff will find workarounds on personal phones, where no audit can see them. For anyone whose tax records or case files sit in these systems, that’s the risk worth watching.