Saturday, 31 October 2026Global tech news & guides — what changed, and what it means for you.
AuTechWire
Original editorial illustration of a calm government office at dusk where a row of work computers shows a single approved assistant window glowing softly, while two other browser windows fade to grey behind a thin security gate

Sweden’s Government Blocks ChatGPT and Claude on Work Computers — but Approves Copilot

by

Sweden’s Government Blocks ChatGPT and Claude on Work Computers — but Approves Copilot

In brief

  • From Oct 26, Sweden’s Government Offices blocked ChatGPT and Claude on staff computers, directing staff to Copilot. The models are not gone — the browser route is.
  • Staff are directed instead to Microsoft Copilot, which the Government Offices reviewed and then tightened the security settings on before making it available. The specific tightened settings were not disclosed, for security reasons.
  • The reason given is information security, not model quality: digitalisation director Mikael Wettercrantz warned on the staff intranet that using AI services without a formal agreement can have consequences for information security, citing examples of problematic use of open AI services.
  • This is not a ban on the models themselves. Staff can still reach ChatGPT and Claude language models through Intric, an internal AI assistant platform — what is blocked is the consumer-style route of opening a chatbot in a browser outside any contract the institution has negotiated.
  • Sweden’s Tax Agency and the National Board of Institutional Care already have similar blocks in place, and the Enforcement Authority is preparing its own decision. The Government Offices had already restricted DeepSeek.

What actually changed on Monday

The reporting is precise about the scope, and the precision is the story. Sveriges Radio’s Ekot first reported the decision on October 23, 2026, and the Government Offices confirmed it in an email to broadcaster SVT: “The browser interface for the AI services ChatGPT and Claude will no longer be reachable from the Government Offices’ work computers from October 26.”

Three limits in that sentence matter. First, it is the browser interface that is blocked, not the models. Second, it is the Government Offices’ work computers — the prime minister’s office and the ministries — not Sweden as a country, and not personal devices. Third, it starts on a stated date, October 26, which is why the date matters: staff arriving at work on Monday found a tool they used on Friday no longer opening.

The approved alternative is Microsoft Copilot, which all staff can use. The Government Offices says it reviewed Copilot and raised its security settings after that review. It has not said what the changes were, and the press office declined to give details, citing security. That silence is normal for a security review, but it leaves one question open that a careful article should not fill in: whether the Copilot available to staff includes Anthropic models. In the EU, Anthropic models in Copilot are off by default and must be turned on by administrators, and the Government Offices has not said whether it enabled them.

Why the contract matters more than the model

Wettercrantz’s intranet warning, as relayed by multiple outlets, is about agreements, not about whether ChatGPT or Claude are good tools. Using AI services without a formal agreement, he wrote, can have consequences for information security. The problematic uses he cited included sensitive information being entered into open systems and AI-generated answers being handled improperly in work.

That framing explains the apparent contradiction at the heart of the decision: the same models that are blocked in a browser remain available through Intric. Intric is an internal AI assistant platform supplied under contract — Swedish AI company Intric won the contract to supply the platform in the summer of 2026, according to one account — and access through it happens under the Government Offices’ own terms and controls. The browser route has no such contract: queries go to servers operated by outside companies, under those companies’ terms, with no institutional control over what happens to staff inputs.

This is the same distinction other governments have been drawing, and Sweden is not the first mover inside its own administration. The Tax Agency and the National Board of Institutional Care, which runs youth homes and addiction treatment facilities, already block these services. The Enforcement Authority, which collects debts and enforces court orders, is preparing a decision. A separate internal audit at the Transport Administration found around ten AI services in widespread and risky use. Outside Sweden, the European Parliament disabled AI features on work devices in February 2026 because it could not guarantee data security. And in the United States, the Department of Homeland Security in October 2026 required civilian federal agencies to remove DeepSeek products from their information systems — a different tool, but the same procurement-and-control logic.

What this means for other organisations

The Sweden case is a template, and it is worth reading as one, because the same question is now landing on every large employer’s desk: staff already use AI tools, the tools are useful, and the institution has no contract with most of them. Sweden’s answer has four parts, and each part is separable:

  1. Block the unmanaged route, not the capability. Staff lose the browser shortcut, not access to AI assistance. That makes the policy enforceable without making the work impossible.
  2. Approve one tool under contract, and say what was reviewed. Copilot passed a review that raised its security settings. The review, not the brand, is the approval.
  3. Keep the models reachable through a controlled platform. Intric preserves the capability inside institutional controls, which is why “Sweden bans ChatGPT” is the wrong headline.
  4. Expect the perimeter to spread agency by agency. The Tax Agency and the institutional-care board moved first; the Government Offices followed; the Enforcement Authority is next. A policy that looks like one decision is usually a sequence.

For individual staff, the practical consequence is narrower than the headlines suggest: on a Government Offices computer from October 26, open Copilot or the Intric platform, not chatgpt.com or claude.ai in a browser. For other organisations watching, the question to copy is not “which chatbot do we block?” but “which AI routes do we have a contract and controls for, and which are just a browser tab?” Sweden answered the second question, and the first answer followed from it.

What to watch next

Two things would change this story, and both are worth checking before this article is treated as settled. First, whether the Government Offices clarifies the Intric scope — which staff keep access, to which models, and under what restrictions — since current reporting says only that part of the staff will keep that access. Second, whether the Enforcement Authority’s expected decision follows the same template, and whether other ministries or agencies publish their own versions. If the pattern holds, Sweden will end up not with one AI policy but with a set of agency policies that share the same contract test — and that test, rather than any single block, is the development to track.

Related reading

Sources

Last verified: October 27, 2026.

Shoaib Hassan

Shoaib Hassan is the founder and editor of AuTechWire, an independent technology publication covering broadband, mobile networks, privacy, startups, and consumer tech.

Related Stories

Leave a Comment