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

For two years, the quiet reality in many offices has been the same: people open a chatbot in a browser tab, paste in a draft, and ask for help. Starting today, that no longer works inside the heart of the Swedish state. From Monday, 26 October 2026, the browser versions of OpenAI's ChatGPT and Anthropic's Claude can no longer be reached from the work computers of Sweden's Government Offices, known in Swedish as Regeringskansliet. That is the body that supports the Prime Minister's Office and the ministries.

The news was first reported on Friday, 23 October, by Ekot, the news desk of the public broadcaster Sveriges Radio. The Government Offices then confirmed the change in an email to SVT, Sweden's public television broadcaster. Staff are being pointed to Microsoft Copilot instead. The reason given is information security, not a verdict on how good the chatbots are.

The three things you need to know

1. The switch. From 26 October, the browser interfaces of ChatGPT and Claude are no longer reachable from staff work computers at the Government Offices. In its email to SVT, the Government Offices wrote that the browser interface for the two AI services "will no longer be reachable" from its work computers from that date. Staff are directed to the AI tools the organization has approved, including Microsoft Copilot. According to internal information seen by Ekot, the Government Offices also raised the security level of Copilot after a review. The press office would not say what exactly was tightened, because the details concern security.

2. The reason. Mikael Wettercrantz, the Government Offices' director of digitalisation, declined an interview with Ekot. In a comment on the staff intranet he wrote that using AI services without an agreement can have "consequences for our information security". The block is not a total ban on the underlying models. According to the Government Offices, staff still have access to the ChatGPT and Claude language models through Intric, an internal AI assistant platform. Ekot's report says some staff will keep that access through a platform the Government Offices developed for AI assistants.

3. It is spreading. The Government Offices is not alone. Ekot's survey found that the Swedish Tax Agency (Skatteverket) and the National Board of Institutional Care (Statens institutionsstyrelse, SiS) have already introduced similar blocks, and the Swedish Enforcement Authority (Kronofogden) is preparing its own decision. At the Swedish Transport Administration (Trafikverket), an internal audit earlier this year found widespread and risky use of about ten different AI services.

What exactly changes on Monday

The wording matters here, so it is worth being precise. What the Government Offices describes is a block of the web interfaces of two consumer AI services on its own work computers. Nothing in the reports says that ChatGPT or Claude are banned from Swedish public life, and nothing says that individual civil servants are forbidden from using AI at all. The opposite is true: the Government Offices told SVT that all staff have access to Microsoft's AI tool.

The block also does not cut off the models themselves. Through Intric, the internal platform, staff can still reach the ChatGPT and Claude language models. That is an important detail. The problem the Government Offices is trying to solve is not "these models are dangerous". It is "we do not control where the information goes when people use the public websites".

There are also things the reports do not cover. The Government Offices has not said how many employees are affected, how the internal platform is configured, or whether the Anthropic models that Microsoft offers inside Copilot are switched on in its own setup. Notebookcheck points out that in the EU those Anthropic models are off by default in Copilot and have to be enabled by administrators. So the line between "blocked" and "approved" is mostly about contracts and control, not about which company trained the model.

Why "without a contract" is the key phrase

Wettercrantz's intranet comment points to the real issue. When an employee types into a consumer chatbot, the text goes to the provider's servers under the provider's standard terms for that kind of account. A government organization that has no agreement with the provider has very little say over how long that data is kept, where it is processed, and who can access it later.

An enterprise agreement changes that. It can set retention periods, data location, and whether inputs may be used to train future models. It also gives the organization logs and admin controls, so it can see what is being used and switch features on or off. Copilot passed the Government Offices' review because the organization has a contract with Microsoft and could raise the security settings itself. For ChatGPT and Claude in the browser, it apparently had neither.

The Transport Administration audit shows why this is not a theoretical worry. Its internal auditors wrote that they had seen "several examples of problematic use of open AI services", concerning both how sensitive the information entered was and how the AI generated answers were handled. That second part is easy to overlook. Information security is not only about what goes into a chatbot. It is also about whether an AI answer ends up in an official document without anyone checking it.

Part of a wider pattern

Sweden's move fits into a pattern that has been building across European institutions. The Government Offices had already restricted the Chinese AI service DeepSeek, according to Ekot. Notebookcheck notes that the European Parliament disabled AI features on work devices in February because it could not guarantee data security. Inside Sweden, at least two agencies blocked consumer chatbots before the Government Offices did, and a third is preparing its decision.

What is new is the target. Blocking a Chinese chatbot over data concerns has become routine for many governments. Blocking the two most popular Western chatbots at the center of a European government is a stronger signal. It says that the issue is not where a company is based but whether the organization has a contract and real control over the data.

For OpenAI and Anthropic, this is also a sales story. Both companies sell enterprise and government products with stricter data terms than their consumer apps. A decision like this one tells every public body in Europe that the free or personal tier is not acceptable for official work, and that the route in is a negotiated agreement, whether directly or through a partner such as Microsoft.

Why it matters

There are three reasons this small IT change deserves attention beyond Sweden.

First, it is a clear example of "shadow AI" being pulled back under control. In many organizations, employees adopted chatbots faster than their IT and legal teams could write rules. The Swedish Government Offices is now drawing a simple line: approved tools with contracts are fine, unapproved consumer tools are blocked on work computers.

Second, it shows that governments are separating models from products. The same ChatGPT and Claude models remain available inside an internal platform. What is blocked is the public website and its default data handling. That is a more nuanced position than a ban, and it is likely to be copied.

Third, it raises a question every company should ask itself. If a government with strict rules on public records decides that pasting work into a public chatbot is a security risk, what does that say about the many businesses that still have no AI policy at all?

Dany's take

I think this is the right call, and I say that as someone who uses ChatGPT and Claude every day. The block is not anti AI. Staff keep an approved assistant and still reach the same models through an internal platform. What disappears is the easy, invisible path where a civil servant pastes a sensitive draft into a public website and nobody knows where it ends up.

The part I would watch closely is the gap between policy and habit. People do not stop using tools that save them time. If the approved option feels slower or weaker, some will move to their phones or private laptops, and the risk simply leaves the network where IT could see it. The real test for the Government Offices is whether Copilot and the internal platform are good enough that nobody misses the browser tab.

My practical advice for anyone running a team: do not wait for an incident. Write down which AI tools are approved, make sure there is a contract behind each one, and give people a good option so they do not need a bad one.

What to watch next

The Enforcement Authority is preparing its own decision, so the list of Swedish agencies with blocks may grow soon. It will also be worth watching whether the Government Offices explains how many staff use the internal platform, and whether other Nordic and EU governments follow with similar rules for consumer chatbots.

Should your workplace ban consumer AI chatbots, or is a clear policy with approved tools enough? Tell us what you think on our social channels.

Sources

• Sveriges Radio Ekot: Regeringskansliet spärrar AI-tjänsterna Chat GPT och Claude
• SVT Nyheter: AI-tjänster spärras på Regeringskansliet
• Decrypt: Sweden's Government Offices Block ChatGPT and Claude, Keep Microsoft Copilot
• TechRadar: ChatGPT and Claude blocked by Swedish government over "problematic" security concerns
• Notebookcheck: Sweden's government blocks ChatGPT and Claude, keeps Copilot