The logos of OpenAI’s ChatGPT and Anthropic’s Claude (illustrative). Image: OpenAI logo (LobeHub Icons, MIT), Claude logo (Simple Icons, CC0); trademarks of their owners

Civil servants in Sweden’s government will no longer be able to open ChatGPT or Claude on their work computers from Monday. The Government Offices (Regeringskansliet), which runs the prime minister’s office and every ministry, is blocking both chatbots on staff machines and pointing employees to Microsoft’s Copilot instead, Sveriges Radio’s news programme Ekot reported on Friday.

The block applies to the computers staff use for work, and it isn’t a ban on AI. Employees are being told to stick to the AI services the Government Offices has approved, and Copilot is one of them. According to internal information Ekot has seen, the Government Offices also tightened Copilot’s security after reviewing the tool. Its press office wouldn’t go into detail, saying the matter concerns security.

“Consequences for our information security”

The reasoning comes from Mikael Wettercrantz, the Government Offices’ director of digitalisation. He turned down an interview with Ekot, but in a post on the staff intranet he warned that using AI services without a contract could have “consequences for our information security”.

That is the line being drawn: Copilot is a service the Government Offices has approved, while staff opening ChatGPT or Claude on the web were using them with no agreement in place. Some will keep both chatbots anyway. The press office told Ekot that certain employees can still reach ChatGPT and Claude through an AI-assistant platform the Government Offices has built itself, without saying who or how many. Use of China’s DeepSeek, which many governments now bar from official devices, was already restricted.

Neither OpenAI nor Anthropic has commented publicly.

Copilot can run Claude too

There is a twist. Copilot isn’t only Microsoft’s own AI. Microsoft’s documentation says Copilot can use Anthropic’s Claude models in features such as Researcher, Copilot Studio and the Copilot tools in Word, Excel and PowerPoint, with Anthropic working as a Microsoft subprocessor under Microsoft’s contract terms.

Those Claude models are excluded from Microsoft’s EU Data Boundary, its commitment to process European customers’ data inside the EU. For organisations in the EU, the EFTA countries and the UK, the main setting that lets staff use Anthropic’s models is off by default, so administrators decide. The Government Offices hasn’t said whether Claude is switched on inside its Copilot.

Not the first Swedish agency to do it

The Swedish Tax Agency (Skatteverket) and the National Board of Institutional Care (SiS), which runs secure youth homes and addiction treatment centres, have already brought in similar blocks, and the Swedish Enforcement Authority (Kronofogden) is considering one, according to Ekot. Elsewhere governments are moving the other way: the US health department says only Claude will do for its staff.

The decision also lands awkwardly for the man at the top. Prime Minister Ulf Kristersson came under fire in August 2025 after telling the business paper Dagens industri that he used ChatGPT and Mistral’s Le Chat for a second opinion:

I use it myself quite often. If for nothing else than for a second opinion. What have others done? And should we think the complete opposite?

Ulf Kristersson, Sweden’s prime minister, to Dagens industri

His spokesperson, Tom Samuelsson, said at the time that “it is not security sensitive information that ends up there”. Virginia Dignum, a professor of responsible AI at Umeå University, put the objection more bluntly to Dagens Nyheter:

We must demand that reliability can be guaranteed. We didn’t vote for ChatGPT.

Virginia Dignum, Umeå University, to Dagens Nyheter

Ekot’s report doesn’t say whether the prime minister’s own use is affected.

Why it matters

Governments are starting to choose AI by contract rather than by brand, and the chatbots staff reach for first are being swapped for whatever comes with an agreement attached. It also shows how thin the line can be: the same data rules that separate one chatbot from another decide whether Claude reaches Swedish officials through Copilot instead.

Sources: Sveriges Radio Ekot; Microsoft Learn: Anthropic models in Microsoft Online Services; The Guardian (August 2025).

Latest Policy & Safety news

More Policy & Safety news