- Lead. Sweden’s Government Offices — the body that runs the prime minister’s office and every ministry — will block civil servants from accessing ChatGPT and Claude on their work computers from Monday, October 26, 2026.
- Fact. Microsoft Copilot is the sole approved alternative, having passed the Government Offices’ internal security review; Sweden’s Tax Agency and the National Board of Institutional Care have imposed similar bans already.
- Stake. The decision joins a growing pattern of European government AI governance, and raises an unresolved question: Microsoft has integrated Claude into Copilot since early 2026, but Sweden’s Government Offices has not confirmed whether that integration is switched on in its own deployment.
The directive was posted to the staff intranet by Mikael Wettercrantz, Sweden’s director of digitalisation, and first reported by Sveriges Radio’s Ekot on October 23. Wettercrantz cited “consequences for our information security” if AI services are used without a formal contract in place. His notice also referenced “problematic use of open AI services” — specifically the sensitivity of information civil servants were entering into consumer AI tools and the way AI-generated outputs were being handled inside government workflows.
Why Copilot Passed, and the Others Did Not
The Government Offices’ press office declined to give specific details of what distinguished Copilot from its rivals in the review, citing security concerns. What is public is that the Government Offices reviewed Copilot and subsequently raised the security settings on its deployment before approving it. ChatGPT and Claude received no equivalent review.
A structural factor that the directive notes, without elaborating on, is US law: consumer AI services send queries to external servers, and US legislation enables authorities to compel those companies to hand over data. For a body handling confidential government information, that creates a risk profile that a locally negotiated enterprise contract — with audit rights and data-processing terms — can at least partially mitigate. The EU AI Office opened formal cases against both OpenAI and Anthropic this year, adding regulatory uncertainty to the mix.
A Broader European Pattern
Sweden’s Government Offices is not acting in isolation. Sweden’s Tax Agency and the National Board of Institutional Care have already imposed comparable restrictions. The Enforcement Authority is preparing its own decision. The European Parliament disabled AI features for its staff in February — including features from ChatGPT, Claude, and even Copilot — a stricter approach than the one Sweden adopted. An internal audit of Sweden’s Transport Administration found approximately ten AI services being used in what the report described as a “widespread and risky way.”
Some staff inside the Government Offices will retain access to an internally built AI assistant platform through which they can reach both ChatGPT and Claude — though the Government Offices has not described how many civil servants that covers or what data-handling restrictions apply. The effective line between banned and approved tools may therefore be narrower in practice than the public announcement suggests.