Sweden’s Government Offices cut off ChatGPT and Claude on staff PCs

Sweden’s Government Offices (Regeringskansliet) will block ChatGPT and Claude on staff work computers from Monday, according to Swedish media reports. Employees are to use Microsoft Copilot instead. The move was first flagged in AI communities on Reddit and traced to reporting by Sveriges Radio’s Ekot.

The Reddit thread is only a pointer. The Government Offices has not, as far as we could find, published a public statement. Everything below comes from Swedish outlets that cite Ekot, including Computer Sweden, and none of it has been confirmed by an independent source.

An intranet warning about services without a contract

According to Computer Sweden’s account of the Ekot report, digitalization director Mikael Wettercrantz wrote in an intranet comment that using AI services without a contract can have consequences for the Government Offices’ information security. The reported rule is simple: staff should use only approved services, and the consumer versions of ChatGPT and Claude are not on that list.

Copilot is the approved alternative. Computer Sweden reports that the Government Offices has also raised the security around it, but the reports we read do not say what was changed. DeepSeek, the Chinese chatbot, had already been restricted earlier, according to the same coverage.

The block covers direct access from work computers. The reports we could read do not say whether it extends to phones, to personal devices used for work, or to API access, and they describe no exceptions.

Other agencies are reportedly doing the same

Ekot also reports that similar blocks have been introduced at the Swedish Tax Agency (Skatteverket) and at the National Board of Institutional Care (Statens institutionsstyrelse, SiS). The Enforcement Authority (Kronofogden) may follow, according to the same report. If accurate, the Government Offices would be part of a pattern in Swedish public administration, not an isolated decision.

The rationale matches what other public bodies have cited elsewhere. A web search for comparable cases turned up reports that the European Parliament’s IT department disabled AI tools, including Claude, Copilot and ChatGPT, on lawmakers’ work devices in February 2026 over data security concerns. In that case the stated worry was also where sensitive text ends up once it is typed into a cloud chatbot.

What changes for officials and for the vendors

For civil servants, the practical change is a switch of tools, not an end to AI at work. Staff who drafted or summarized text in ChatGPT or Claude will have to do it in Copilot, which the Government Offices already has a contract for. A contract matters here because it sets terms on data handling, which a free or personal account does not.

For OpenAI and Anthropic, the effect is a lost foothold in a visible government customer, though the reports do not say that either company was in talks about a contract. Neither company has commented in the coverage we read. The block also favors Microsoft, which is already the incumbent supplier of office software to many public bodies.

The decision sits awkwardly beside Sweden’s recent AI debate. As we reported in August 2025, Prime Minister Ulf Kristersson said he uses AI tools such as ChatGPT as a second opinion, which drew public criticism. His office said at the time that no sensitive information is shared with them. The new rule suggests that the government’s own IT policy is drawing a harder line than that earlier public conversation implied.

What the reports do not establish

Several points remain open. The Government Offices’ press office has reportedly declined to give details, citing security reasons, so the exact scope and any exemptions are unconfirmed. It is also unclear whether the block is permanent or a stopgap until more services are procured and reviewed. And nothing in the reports indicates a specific security incident behind the decision; the stated reason is the lack of a contract.

Readers should treat the details as reported, not official. If the Government Offices publishes its guidance, the questions to check first are whether enterprise versions of ChatGPT or Claude could be approved later, and how the Copilot security changes were verified.