Sweden Blocks ChatGPT and Claude on Government PCs Today

From today, Monday 26 October 2026, staff at Sweden's Government Offices can no longer open the ChatGPT and Claude websites on their work computers. It is not a ban on AI itself, and it is not a new law. It is a workplace IT decision, and it tells us a lot about how governments are starting to draw the line around everyday AI tools.

The block was first reported on 23 October by Ekot, the news desk of Sweden's public broadcaster Sveriges Radio, and later confirmed by the Government Offices in an email to broadcaster SVT. The message was blunt: "The browser interface for the AI services ChatGPT and Claude will no longer be reachable from the Government Offices' work computers from October 26."

If you use ChatGPT or Claude at work, or you run a team that does, this story matters beyond Sweden. The same question is now landing on desks everywhere: which AI tools are actually approved, and what happens to the data you paste into the ones that are not?

Quick answer: what exactly is blocked in Sweden?

From 26 October 2026, the public web versions of OpenAI's ChatGPT and Anthropic's Claude are blocked on staff work computers at Sweden's Government Offices, known in Swedish as Regeringskansliet — the administration that supports the prime minister and the cabinet. Microsoft Copilot stays, because it is an approved service whose security settings were tightened after an internal review. Some staff will still reach ChatGPT and Claude models through a controlled internal assistant platform, rather than through the open websites.

What is Sweden's Government Offices block, in plain terms?

Think of it less like a country banning a website and more like an office locking down its own computers. The Government Offices are not blocking ChatGPT and Claude for all Swedes. They are blocking the browser versions on their own staff machines, so employees use only tools the organisation has formally approved.

That distinction matters, because early headlines made it sound bigger than it is. Nobody in Sweden has lost access to ChatGPT on their own phone or home computer. Government staff have lost the shortcut of opening the public chatbot sites at their desks — and that is precisely the point. On a work computer, what you paste in can be work product, citizen data, or a draft policy. The office wants to know where that text goes.

Who reported the block, and when?

Ekot reported the decision on 23 October 2026, citing internal information it had seen. The Government Offices then confirmed the block to SVT by email, giving the start date as Monday 26 October. Digitalisation director Mikael Wettercrantz declined an interview with Ekot, but wrote on the offices' intranet that using AI services without an agreement can have "consequences for our information security."

That last phrase is the whole policy in one sentence. The problem, as the offices frame it, is not that ChatGPT or Claude are bad tools. It is that there was no contract governing them — no agreed rules about data handling, retention, and access — and in government work, that gap alone is a security issue.

Why did Sweden block ChatGPT and Claude?

Three reasons keep coming up in the reporting, and they are worth separating because they point to different fixes.

1. No contract, no control over the data

When you type into a public chatbot, your prompt leaves your computer and is processed on the provider's servers. Without a formal agreement, the organisation has limited say over how long that data is kept, who can access it, and under which country's laws it can be requested. Reporting on the decision notes that US law can compel companies to hand over data to authorities — a real concern for a government handling sensitive material.

This is also why Copilot survived the cut. Microsoft Copilot is an approved service at the Government Offices, and internal documents seen by Ekot say its security settings were raised after a review. Approval, in other words, is not a popularity contest. It is a contract plus a configuration.

2. Staff were already using open AI tools in risky ways

The block did not come out of nowhere. Sweden's Tax Agency (Skatteverket) and the National Board of Institutional Care (SiS) have already introduced similar restrictions, and the Swedish Enforcement Authority (Kronofogden) is weighing a decision. An internal audit at the Swedish Transport Administration found staff using about 10 different open AI services widely and at high risk, including cases where sensitive information was entered and AI-generated answers were mishandled.

None of the reporting on the Government Offices block says a specific leak triggered it, and that is an important honesty check. The stated logic is the absence of a contract and a pattern of risky use across agencies — not a documented breach at the Government Offices themselves.

3. DeepSeek was already restricted

The Government Offices had already restricted the Chinese AI service DeepSeek before this move, so ChatGPT and Claude are not the first tools to hit the wall. If you want the backstory on why DeepSeek keeps getting this treatment from institutions, our DeepSeek V4.1 explainer covers what the model is and why organisations treat it differently from Western frontier models.

What stays available: Copilot, and the internal platform

Here is what staff can still use, and why each option survived:

Microsoft Copilot — the approved assistant for all staff, with security settings tightened after an internal review. If your organisation has a Microsoft agreement, this is the closest parallel to what Sweden did.
A controlled internal assistant platform — some staff will still reach ChatGPT and Claude models through this platform, inside the organisation's own controls, instead of through the public websites. The models are not gone; the uncontrolled doorway to them is.
Personal devices, outside work — the block applies to Government Offices work computers, not to private use. That boundary is exactly what makes this an IT policy story rather than a censorship story.

The lesson for any workplace is simple: banning the website is easy, but the useful version of this policy is approving a safe path and making it good enough that staff actually use it. A block with no approved alternative just pushes people to their phones.

Blocked vs approved: the Sweden comparison

Tool / route	Status on Government Offices work PCs	Why
ChatGPT website (OpenAI)	Blocked from 26 Oct 2026	No formal agreement; public web interface
Claude website (Anthropic)	Blocked from 26 Oct 2026	No formal agreement; public web interface
Microsoft Copilot	Approved, security settings raised	Approved service, passed internal review
Internal assistant platform	Available to some staff	Controlled route to models, inside office controls
DeepSeek	Already restricted	Earlier restriction, before this decision

Notice the pattern: the dividing line is not "good AI vs bad AI." It is "contracted and configured vs open and uncontracted." That is also why this story connects to the wider agent wave — Anthropic's Claude Duet agent app and OpenAI's GPT-6 intelligent interface are both pushing AI deeper into daily work, which makes the question of approved vs unapproved access more urgent, not less.

Is this part of a bigger European trend?

Yes, and Sweden is only the latest example. In the same week, Italy's Legislative Decree 179/2026 took effect on 23 October 2026, requiring a real person to make final decisions on hiring, discipline, and dismissal — a decision resting solely on automated processing would be void. Italy chose a labour-law tool; Sweden chose an IT-security tool. Both are about the same thing: keeping a human and an institution in control when AI is part of the job.

Zoom out further and the same week also brought the UK's announcement of a new AI Biosecurity Lab, backed by £100 million of government funding plus up to £200 million in model credits and support from OpenAI and Anthropic. Governments are not turning away from AI. They are building approved, governed channels for it — labs, contracts, and internal platforms — while closing the ungoverned ones.

What this means for you: a 5-step workplace AI check

Whether you work in an office, run a small team, or freelance for clients, here is the practical version of Sweden's decision, as a checklist you can actually use:

Find out what is approved. Ask IT, or your client, which AI tools have a contract behind them. If nobody can answer, assume the public chatbot is not approved for sensitive work.
Never paste what you could not email. A simple rule that covers most accidents: if a document is too sensitive to send to an outside company by email, do not paste it into a public chatbot either.
Use the approved tool first. In Sweden's case that is Copilot. In your workplace it might be an enterprise plan of ChatGPT, Claude, or Gemini, or an internal tool like the rebuilt Notion AI agents setup many teams now run inside their workspace. The approved tool is the one with the agreement.
Treat AI output as a draft, not a decision. Check facts, check names, check numbers. Sweden's Transport Administration audit specifically flagged mishandled AI-generated answers — the output is where the second half of the risk lives.
Keep a personal-device boundary. Do work AI on work-approved tools, and keep personal AI use personal. Mixing the two is how work data ends up in a consumer account with consumer terms.

Pros and cons of Sweden's approach

What works

It closes a real data-control gap instead of pretending the gap does not exist.
It keeps an approved alternative (Copilot) rather than leaving staff with nothing.
It preserves model access through a controlled platform, so capability is not lost.
It sends a clear signal to other agencies: get a contract, or get blocked.

What to watch

Staff may simply switch to personal phones, moving the risk rather than removing it.
Copilot's tightened settings are not public, so outsiders cannot judge how strong the approved route really is.
Two-tier access — some staff on the internal platform, others not — can create confusion about who may use what.
A block is not a strategy on its own; without training, the next risky paste will happen in the next approved tool.

Our verdict

Sweden's block is not anti-AI, and it is not the start of a European chatbot ban. It is a procurement and security decision dressed up as a headline: no contract, no access on government machines. For readers, the useful takeaway is not "ChatGPT is in trouble" — it is not — but "the free public version and the contracted enterprise version are now different products in the eyes of institutions, and you should know which one you are using."

If your workplace has not had this conversation yet, Sweden just had it for you. Ask the approved-tool question before an incident asks it for you.

Covering this story as a video? If you make an explainer for YouTube, you can grab a clean high-resolution thumbnail for any public video with our free YouTube Thumbnail Downloader — handy when you need a reference image for your own edit.

Frequently asked questions

Is ChatGPT banned in Sweden?

No. Only the ChatGPT and Claude websites are blocked on work computers at Sweden's Government Offices from 26 October 2026. Private use in Sweden is unaffected, and the block is an internal IT decision, not a national law.

Why did Sweden block ChatGPT and Claude?

The Government Offices said using AI services without a formal agreement can have consequences for information security. Staff are directed to approved tools instead, chiefly Microsoft Copilot, whose security settings were tightened after a review.

Can Swedish government staff still use AI at all?

Yes. Microsoft Copilot remains available to all staff, and some staff can still reach ChatGPT and Claude models through a controlled internal assistant platform rather than the public websites.

Did a data leak cause the Sweden ChatGPT block?

No public statement says a specific leak at the Government Offices triggered the block. Reporting points to the lack of a contract and a pattern of risky open-AI use found in audits at other Swedish agencies.

Which other Swedish agencies have similar AI restrictions?

The Tax Agency and the National Board of Institutional Care have introduced similar restrictions, the Enforcement Authority is weighing a decision, and DeepSeek was already restricted at the Government Offices before this move.

Will other countries follow Sweden and block ChatGPT at work?

Workplace restrictions are already spreading agency by agency, but a full national ban looks unlikely. The more probable pattern is Sweden's: public chatbot sites blocked on work machines, with contracted, approved AI tools offered instead.

Last updated: 26 October 2026. How we reported this: we checked the Ekot report of 23 October 2026 as carried by multiple outlets, the Government Offices' confirmation to SVT, and the wider European context, and we separated confirmed facts from interpretation where reporting differed. This article is not sponsored, and no AI company paid for coverage.

Search This Blog

The AI Tools