Sweden Just Blocked ChatGPT and Claude on Government Computers. Here's Why Copilot Stayed

Starting Monday, October 26, 2026, staff at Sweden's Government Offices (Regeringskansliet) can no longer open ChatGPT or Claude in their browsers on work computers. The Government Offices include the Prime Minister's Office and every ministry. Microsoft Copilot remains available. The office says it reviewed Copilot and tightened its security settings before keeping it. The story was first reported by Sveriges Radio's Ekot, and the office later confirmed the block to SVT by email.

The official reason is information security. Using AI services without a formal agreement, the office says, can put government information at risk. Mikael Wettercrantz, the office's director of digitalization, told Decrypt the office had seen several examples of problematic use of open AI services. The office hasn't shared details of those cases.

Yes, indirectly. According to SVT, staff can still reach OpenAI and Anthropic models through Intric, an internal AI platform the government controls. That's the pattern to notice: the models are fine, the uncontrolled access route isn't. The office hasn't said whether Anthropic's Claude models are turned on inside Copilot. In the EU they're off by default, and an administrator has to switch them on.

No. Several Swedish agencies already restrict AI chatbots:

About 10: the number of AI services Sweden's Transport Administration found in widespread, risky use during an internal audit.

This is the part most people miss. The same model can be risky or safe depending on how you access it.

Sweden's move is a preview of what auditors and regulators will expect from companies too. Staff are already pasting contracts, customer data and code into free AI tools. If that's happening in your company without an agreement, you have the same risk Sweden just shut down.

• List every AI tool your team actually uses (ask, don't assume)
• Pick approved tools with a business or enterprise agreement
• Turn off model training on company data in each tool's settings
• Write a one-page AI use policy: what data can and can't go in
• Block or discourage unapproved AI sites on work devices
• Review the list every quarter, since new tools appear monthly

In the short term, yes. Copilot comes bundled with Microsoft 365, which most governments already pay for, so the contract and admin controls already exist. That gives Microsoft a strong edge in public-sector AI. But OpenAI and Anthropic both sell enterprise plans with the same kinds of guarantees. The real competition is now about who signs the contracts, not who has the smartest chatbot

Sweden didn't ban AI. It banned AI it couldn't control. Any company letting staff use free AI tools with sensitive data should ask the same question Sweden did: do we have a contract that says where our data goes?