Swedish ministries cut off ChatGPT and Claude but keep Microsoft's Copilot

Starting Monday, October 26, ChatGPT and Claude will no longer open on the work computers of the Regeringskansliet, the Swedish Government Offices, which bring together the Prime Minister's office and the ministries. Staff are being steered toward Copilot, Microsoft's AI assistant.

The news was revealed on October 23 by Ekot, the newsroom of public radio broadcaster Sveriges Radio, based on internal information. The magazine Computer Sweden and public television SVT then picked it up. The Regeringskansliet press office declined to give details, saying the subject touches on security matters.

Why is Copilot still allowed?

The justification appears in a comment published on the intranet. The digital director, Mikael Wettercrantz, warns there that AI services used without a contract can have consequences for information security. He declined the interview Ekot offered him.

Copilot, by contrast, was reviewed, and the service's security was then strengthened, according to the internal information the radio station consulted. What was examined or changed is not specified.

The block is not total, either. According to the press office, some staff will keep access to ChatGPT and Claude through an AI assistant platform developed in-house. That suggests the problem is less the tool than the access channel: a consumer service opened directly in the browser, or access that the administration controls itself.

Other Swedish agencies have already taken the step

This is not a first. The Regeringskansliet had already restricted the use of DeepSeek, the Chinese AI, which Washington has also just banned from its civilian agencies. According to Ekot, the tax agency (Skatteverket) and the public agency in charge of compulsory care institutions (Statens institutionsstyrelse, SiS) have put similar blocks in place. The debt collection authority (Kronofogden) is preparing its own decision.

What a French organization can take from this

There is no vulnerability and no breach in this story: the issue at stake is the contract. A few reflexes apply to any organization whose teams use AI assistants:

• Take stock of actual usage, including personal accounts that employees or civil servants have opened on consumer services.
• Read the terms of use: without an enterprise contract, there is often no guarantee that the data entered will not be retained or reused.
• Offer a managed solution before blocking, like the Swedish in-house platform. A banned tool with no replacement tends to migrate to personal phones.
• Document the review of the chosen tool, Copilot included: an approved service is not automatically a safe one, it is above all one covered by a contract.

One question remains that neither the ministry nor Ekot answers: why did Copilot pass the review, when OpenAI and Anthropic also offer contract-based plans for businesses? The answer will show whether this was a security choice or a vendor choice.