BMShop Products
← All articles
Sweden's Government Offices block ChatGPT and Claude on staff computers
AI News

Sweden's Government Offices block ChatGPT and Claude on staff computers

From 26 October, Sweden's Government Offices block browser access to ChatGPT and Claude on work computers and direct staff to Microsoft Copilot.

Ammar Rahimi · 26 October 2026 · 3 min read

Staff at Sweden's Government Offices, the body supporting the prime minister and the cabinet, can no longer open ChatGPT or Claude from their work computers as of Monday, 26 October 2026. They are directed to Microsoft Copilot, the only approved AI tool. Sveriges Radio, Sweden's national public radio, first reported the decision.

1. What happened

The block cuts off the browser versions of both chatbots on service computers. It does not touch personal phones or home PCs. The Government Offices confirmed the decision in an email to broadcaster SVT.

The instruction came from Mikael Wettercrantz, the offices' director of digitalisation, via the staff intranet. He wrote that using AI services without a formal agreement can have "consequences for our information security".

2. The reason is the contract, not the model

The Government Offices says it reviewed Copilot and tightened its security settings after that review. ChatGPT and Claude did not go through the same process. What separates Copilot is the enterprise agreement and control over its settings, not the model behind it.

Wettercrantz also said his office had seen several examples of problematic use of open AI services. The examples involved both the sensitivity of the information entered and how the AI-generated answers were handled.

3. The interesting exception

This is not a full ban. Some staff can still reach ChatGPT and Claude through Intric, an internal platform the Government Offices built itself and which is used across the ministries. The press office has not said how many people have that access or what restrictions apply.

There is another wrinkle. Microsoft added Claude models into Copilot in early 2026. In the EU, Anthropic's models are off by default and must be switched on by administrators. The Government Offices has not confirmed whether its Copilot instance has Claude enabled. So the line between the "banned" AI and the "approved" one is not as clean as the announcement sounds.

4. Sweden is not the first

Sveriges Radio's reporting shows other agencies moved earlier. Skatteverket, the Swedish Tax Agency, and Statens institutionsstyrelse have already imposed similar blocks. Kronofogden, the Enforcement Authority, is preparing its own decision.

Outside Sweden, the European Parliament went further in February: it disabled AI features on all work devices, including ChatGPT, Claude and Copilot, after concluding it could not guarantee the security of the data those tools processed. Brussels restricted all three, while Sweden approved one and shut out two.

5. Why it matters for small businesses

The decision shows where AI governance is heading: the question is not "which model is best" but "where does our data go". A consumer AI tool sends your queries to an external company's servers under its standard terms. An enterprise contract can set how long data is stored, where it is processed, and whether it can be used for model training. If your business handles customer information, that difference matters.

What it means for you

  1. Check your AI subscriptions. If your work touches sensitive data, use an enterprise account, or at minimum switch off training data in the privacy settings of your consumer account.
  2. Do not assume all tools are equal. Copilot was approved not because its model is better but because a contract and configuration controls exist. Apply the same logic when picking tools for your team.
  3. Plan a route for blocked tools. Sweden kept access through a governed internal platform. Your business can do the same: one approved AI channel instead of a blanket ban.
  4. Assume what you type leaves the building. The problematic use examples in Swedish agencies involved sensitive information entered into open tools. Treat every prompt as data moving to someone else's server.

Sources