Announced 23 Oct 2026 · Sources checked
Sweden's Government Offices are switching off ChatGPT and Claude
Sweden's Government Offices, the Regeringskansliet that supports the prime minister and the ministries, are blocking OpenAI's ChatGPT and Anthropic's Claude on staff work computers. Sveriges Radio's news programme Ekot reported the decision on 23 October 2026, citing internal information it had seen, and the story was quickly picked up by SVT, Computer Sweden and Ny Teknik.
According to Ny Teknik's account of the Ekot report, the block takes effect from Monday, 26 October. From then on, staff will no longer be able to reach the two services from their work machines. Instead they are being directed to Microsoft Copilot, which the Government Offices have approved for use.
The decision is a small, concrete example of a big question every government is working through: which AI tools civil servants may use, on what terms, and who gets to see the text they type into them. Sweden's answer, at least at the centre of government, is to keep AI but limit it to services covered by a formal agreement.
What the block covers
The reports describe a block on access, not a ban on AI. Staff work computers will be stopped from reaching ChatGPT and Claude, which in practice means the consumer and business web versions of the two chatbots used without a contract with the Government Offices. Nothing in the coverage suggests staff are being disciplined for past use, or that the rule applies to private devices outside work.
The reports do not say whether the block also covers access through programming interfaces, browser extensions or third-party apps built on OpenAI and Anthropic models. That detail matters, because many office tools now call these models in the background. The Government Offices had not published the internal guidance, and SVT said it was seeking comment from Regeringskansliet.
It is also not the first restriction. Computer Sweden reports that the Government Offices had already limited the use of DeepSeek, the Chinese AI service that many governments have restricted over data protection and security concerns. The new step extends that approach from a Chinese provider to the two of the best-known American chatbots.
Why officials say they are doing it
The stated reason is information security. Writing on the Government Offices intranet, digitalisation director Mikael Wettercrantz said that using AI services without an agreement can have consequences for our information security, according to the quotes carried by Computer Sweden and Ny Teknik. The block is meant to make sure staff use only the services the Government Offices have approved.
The underlying worry is straightforward. When an official pastes a draft memo, a briefing or an email into a public chatbot, that text leaves the organisation and is processed under the provider's standard terms rather than terms negotiated by the government. Depending on the product and settings, those terms can allow the provider to retain the data or, for some consumer plans, to use it to improve models. For a body that handles policy drafts, negotiations and classified material, that is a real risk even if the provider never misuses the data.
An agreement changes the picture because it can fix where data is stored, how long it is kept, who can access it and whether it can be used for training. That is the same logic behind policies in other countries, such as the US decision to order federal civilian agencies to remove DeepSeek, although the Swedish move is about the absence of a contract rather than about the provider's home country.
Copilot is the approved alternative
The approved tool is Microsoft Copilot. SVT reports that the Government Offices approved Copilot after reviewing the service and raising its security, and Computer Sweden adds that the security of the tool was strengthened as part of that review. The reports did not specify which Copilot product or which settings were changed.
Choosing Copilot fits how most governments already run their offices. Many public bodies have a contract with Microsoft for email, documents and identity, so an AI assistant inside that environment can inherit existing data protection terms, access controls and audit logging. For a security team it is far easier to approve one more feature from an existing supplier than to sign new agreements with several AI labs.
The trade-off is concentration. Steering a whole administration to a single vendor deepens dependence on Microsoft at a time when regulators are scrutinising its position, as in the UK competition regulator's proposal to give Microsoft strategic market status over business software and Copilot. Staff also lose the choice of tool, and ChatGPT and Claude are often preferred for writing and analysis tasks.
Other Swedish agencies got there first
The Government Offices are following other Swedish agencies rather than leading them. According to Ekot's survey, as reported by SVT, Computer Sweden and Ny Teknik, the Swedish Tax Agency, Skatteverket, and the National Board of Institutional Care, Statens institutionsstyrelse or SiS, have already introduced similar blocks.
The Swedish Enforcement Authority, Kronofogden, is currently preparing the question, the reports say. That suggests a pattern of agencies with sensitive personal data moving first: tax records, people in compulsory care and debt enforcement files are exactly the kinds of information that should not end up in a public chatbot.
Taken together, the reports describe a Swedish public sector converging on a common rule: AI is allowed when it runs under a contract the agency controls, and blocked when it does not. That is a more workable line than a blanket ban, but it puts the burden on each agency to review and approve tools one by one.
A loophole for some staff
There is an important exception. According to Ekot, as reported by Ny Teknik, some Government Offices staff will still be able to use both ChatGPT and Claude through a dedicated platform that the Government Offices have built themselves.
The reports do not explain how that platform works, but the arrangement points to the real issue being control rather than the models. A platform run by the Government Offices can decide which staff get access, log what is sent, filter sensitive content and connect to the models under negotiated terms. The same model can be acceptable through an internal gateway and unacceptable through a public website.
For staff, this means the block is less about losing access to particular models and more about routing all use through approved channels. For outsiders, it means the headline that the Swedish government has banned ChatGPT and Claude is not quite accurate: it has banned uncontrolled use of them.
Part of a wider pattern in government
Governments are moving in two directions at once. They are signing deals to bring AI into their work, as with Google's Gemini for Government services for US agencies, while tightening rules on tools that staff use without permission. Security teams call this unsanctioned use shadow AI, and it is hard to stop once people find the tools useful.
Europe adds its own pressure. Public bodies handle large amounts of personal data under the GDPR, and the rules for what AI providers may do with that data are still being argued over, as in the recent push by EU governments for a GDPR change on AI training. Until those rules settle, many agencies will prefer tools covered by contracts their lawyers have already reviewed.
For the Government Offices, a clear list of approved services also answers a practical question that otherwise lands on individual officials: whether a given chatbot is safe for a given document. Under the new rule, the answer no longer depends on each person reading a provider's privacy terms.
What it means for AI vendors and civil servants
For OpenAI and Anthropic, the message is that being popular with individual civil servants is not enough to stay on government networks. Both companies sell enterprise and government versions with contractual data protections, and decisions like this one push agencies toward signing those deals or toward incumbents like Microsoft that already have them.
For civil servants, the change mostly means working inside Copilot or the internal platform rather than a familiar chatbot tab. Whether that slows anyone down will depend on how capable the approved tools are and how many staff get access to the internal gateway, details the Government Offices have not published.
The open questions are the scope and the follow-through: whether the rule covers AI features built into other software, whether more Swedish agencies follow Skatteverket and SiS, and whether the Government Offices publish the criteria a service has to meet to be approved. Those criteria would tell AI vendors far more about how to win government business than any single block.
Common questions
Has Sweden banned ChatGPT?
No. The Government Offices are blocking ChatGPT and Claude on staff work computers. The public is unaffected, and some staff can still use both through an internal platform.
Why are ChatGPT and Claude being blocked?
The digitalisation director wrote that using AI services without an agreement can have consequences for information security. Staff must use approved services such as Copilot.
Which other Swedish agencies block AI chatbots?
According to Sveriges Radio Ekot, the Swedish Tax Agency and the National Board of Institutional Care already have similar blocks, and the Enforcement Authority is considering one.
Sources & further reading
How this story was made
Written by Kristian Kostov with AI assistance and checked against the linked sources. Company performance claims are attributed to the company. Analysis reflects AiLookout’s interpretation; we have not independently tested the products discussed. Cover photography is illustrative and does not depict the specific announcement or product.
Our editorial standards
