Sweden’s government didn’t ban AI. It banned the wrong kind of paperwork. From Monday, October 26, 2026, staff at the Regeringskansliet, the Swedish Government Offices, can no longer open ChatGPT or Claude from work computers, while Microsoft Copilot stays on. The difference isn’t capability. It’s a signed data processing agreement.
Swedish public radio’s Ekot reported the block on October 23, based on internal information. It covers the prime minister’s office and every ministry. The measure is a spärr, a network-level block, and the stated goal is to push employees toward tools the Government Offices have already approved.
What changes on Monday
ChatGPT and Claude will stop responding from workstations. Copilot remains, and per Ekot it got a security tightening after an internal review. The press office won’t say what that involves, citing security.
There’s an exception. Some staff will still reach ChatGPT and Claude through an internal assistant platform. A 2025 presentation shows the Government Offices have run more than thirty assistants since 2023 on the Intric cloud platform, which offers several models, including Claude and GPT. So the models aren’t gone. The uncontrolled front door is.
DeepSeek was restricted earlier. This time the block hits the two most widely used Western chatbots, at the center of the executive.
Why the contract matters more than the model
Digitalization director Mikael Wettercrantz wrote on the intranet that using AI services without an agreement can have ‘consequences for our information security.’ He declined an interview.
The logic is simple. A personal ChatGPT or Claude account runs on the provider’s standard terms. Data stays where the provider decides, for as long as it decides, and depending on settings it can end up in training. An enterprise deal like Microsoft’s comes with a DPA, the document GDPR requires when a vendor processes data for an organization, plus written rules on retention and server location.
The real-world risk is mundane. Someone pastes meeting minutes or a draft resolution into the chatbot they use at home because it’s faster than the official tool. Sweden has seen this already. An internal audit at Trafikverket, the transport agency, found risky use of around ten AI services, including problems with both the sensitivity of inputs and how outputs were handled.
The Kristersson backdrop
In August 2025, Prime Minister Ulf Kristersson told Dagens industri he used ChatGPT and Mistral’s Le Chat for a second opinion. The backlash was loud. Researcher Virginia Dignum’s line, ‘We didn’t vote for ChatGPT,’ made the rounds. His office said no security-sensitive information was entered. Whether the new block applies to his personal use is unclear.
Other agencies are doing the same
Ekot found Skatteverket (tax) and SiS (youth care and addiction treatment) already have blocks. Kronofogden, the enforcement agency, is preparing a decision. Trafikverket has the audit findings. The pattern is consistent across agencies.
Elsewhere the logic differs. The US Department of Homeland Security just banned DeepSeek from civilian federal agencies, a decision tied more to the vendor’s country of origin than to contracts.
The Copilot paradox
Here’s the awkward part. Microsoft’s documentation, updated October 23, says Microsoft 365 Copilot can use Anthropic’s Claude models in Researcher, Copilot Studio and features in Word, Excel and PowerPoint, with Anthropic as a Microsoft subprocessor. In the EU, EFTA countries and the UK, those models are off by default and only an admin can enable them. They’re also excluded from the EU Data Boundary. Since April 3, 2026, European admins can even make Anthropic the default Copilot model.
The Regeringskansliet hasn’t said whether Claude is active in its Copilot. Either way, the same model blocked at Anthropic’s site could be running inside an approved Microsoft wrapper.
Does it matter?
Yes, mostly for vendors. OpenAI and Anthropic haven’t commented, but this is the procurement reality they face: governments will pay for AI, just not through a browser tab. Microsoft’s advantage is that it already holds the contract. For founders selling into the public sector, the lesson is blunt. Capability gets you a demo. A DPA gets you deployed.
