Sweden’s Government Offices block ChatGPT and Claude, keep only Copilot

Key Points

1. Government Offices will cut browser access to ChatGPT and Claude from Monday, 26 October 2026.
2. Microsoft Copilot is the sole approved AI tool after a dedicated security review raised its settings.
3. Swedish tax, enforcement and care agencies are moving the same way, setting a public-sector precedent.

The latest:

Browser access to ChatGPT and Claude disappears from work computers at Sweden’s Government Offices on Monday, 26 October 2026, the body confirmed to SVT by email. Microsoft Copilot remains the only approved AI tool, cleared after a dedicated security review that ended with its security settings tightened. Staff keep access to both blocked models through Intric, an internally built AI platform.

Details:

• The decision: The Government Offices — the administrative apparatus serving the prime minister and cabinet — told SVT in an email, with no spokesperson named, that the browser interface for the two AI services would no longer be available. The email was the only confirmation; no public statement was issued.
• The exception: The same email said employees still reach the ChatGPT and Claude language models inside Intric, the platform the Government Offices built in-house. The press office did not say how many staff that covers, nor what restrictions apply to their use of the models.
• The approved tool: Copilot was described as the only AI tool cleared for use, following a security review carried out specifically for it. Its security settings were raised after that review. According to the Government Offices, every employee has access to the Microsoft tool.
• The reasoning: Mikael Wettercrantz, digitalisation director at the Government Offices, set out the justification in a post on the staff intranet, according to Sveriges Radio Ekot as relayed by Decrypt. He said using AI services without a formal contract could carry consequences for the organisation’s information security.
• The incidents: Wettercrantz cited numerous examples of problematic use of open AI services, according to the Ekot report. The problems concerned the sensitivity of information staff were entering into the services and how they handled the answers the AI systems produced. No individual case was described.
• Who else: Sweden’s Tax Agency (Skatteverket) and the National Board of Institutional Care (SiS) have already imposed comparable blocks, SVT reported. The Enforcement Authority (Kronofogden) is weighing a similar decision, leaving at least four national bodies converging on the same restriction.
• The chain: Sveriges Radio Ekot broke the story, a point SVT credited in its own report. SVT published on 23 October 2026 at 12:36 and updated the item on 24 October at 16:21, after confirming the block directly with the Government Offices.
• The timing: The cut-off lands on a Monday, the first working day after the reporting, giving staff no transition window announced publicly. The Government Offices did not set out how long the Copilot-only arrangement will run or whether other vendors could be reviewed later.

Background:

Intric is an AI platform the Government Offices developed internally, which is why the same two models remain reachable there after the public browser versions are cut off. The distinction is contractual and architectural rather than a judgement on the models themselves.

Between the lines:

The dividing line drawn here is not between vendors but between contracted and uncontracted access: the same ChatGPT and Claude models stay available inside Intric while their open browser versions are blocked. Wettercrantz’s stated concern was the absence of a formal contract and what staff were typing into open services, which points to data handling rather than model quality as the trigger.

What’s next

Whether Kronofogden follows through on the block it is considering, and whether the Government Offices publish a public statement, name the restrictions applied inside Intric, or extend the security-review process to any additional vendor.