Sweden’s Government Offices will block ChatGPT and Claude on staff work computers starting Monday, October 26, directing civil servants to approved tools led by Microsoft Copilot, Swedish public broadcaster Sveriges Radio reported on October 23. Digitalisation Director Mikael Wettercrantz told staff on the government’s intranet that using AI services without a data agreement “could have consequences for our information security.” The block covers the prime minister’s office and the central ministries, and other Swedish agencies — including the Tax Agency and the National Board of Institutional Care — have introduced the same restrictions. It is the latest in a string of Western government moves to push official work onto sanctioned AI tools, and some staff will still reach ChatGPT and Claude through a tightly controlled in-house platform.
What The Block Covers
The restriction comes from the Regeringskansliet, the Government Offices that serve the prime minister and the ministries, and takes effect on work computers from Monday. Employees will be directed to approved services, primarily Microsoft Copilot, which the Government Offices cleared after a review that also led officials to strengthen Copilot’s security settings — the details of which they have not disclosed. The decision is not a personal-use ban: staff who need ChatGPT or Claude for their work can still access both through a dedicated AI platform the Government Offices developed in-house, keeping the traffic inside a controlled environment.
The move extends beyond the central offices. Sweden’s Tax Agency and the National Board of Institutional Care have already blocked the services on their own networks, and the Swedish Enforcement Authority is considering the same step. Internal audits showed why: at the Swedish Transport Administration, a review this year found staff using ten different open AI services, with documented cases of employees feeding sensitive information into the systems and mishandling AI-generated answers.
Why Sweden Is Doing This Now
Wettercrantz declined an interview with Sveriges Radio’s Ekot, but the intranet post carrying his warning makes the government’s reasoning plain: AI services used without a contract give officials no say over where the data goes. That is a data-residency and procurement problem rather than a judgment on the models themselves. The Government Offices had already restricted DeepSeek in an earlier decision, and the new block applies the same logic to the two most widely used Western assistants.
The contrast with the United States is instructive. The Department of Homeland Security’s order removing DeepSeek from federal civilian agencies targets a foreign model on supply-chain and security grounds, while Sweden’s block applies to the leading Western assistants on contract and data-handling grounds. Different criteria, same destination: government work consolidates onto the vendor that can sign the security paperwork — and right now, in Sweden, that is Microsoft.
Why It Matters
What makes Sweden’s move worth watching beyond Scandinavia is the method. The Government Offices did not issue a policy memo and hope for compliance; it implemented a network-level block and paired it with a sanctioned alternative that had already passed a security review. That is the complete procurement loop most enterprises are still missing: ban the unsanctioned tools, then actually provide one that passed review, so employees have somewhere legitimate to go.
The decision also redraws the enterprise AI map in one stroke. Copilot is now the only approved assistant for everyday Swedish government work, giving Microsoft’s Copilot platform a government endorsement that its rivals just lost. Expect other governments to watch closely — and vendors to fight harder for the security certifications that decide these competitions.
Background
Governments have been tightening the rules on AI in official work for years, from the EU AI Act’s obligations on public-sector AI use to procurement policies that favor audited tools. Sweden’s approach fits a global pattern of pushing sensitive work toward controlled, contract-backed AI services — the same logic that has driven incident playbooks and procurement standards across the industry, as in the NIST, NSA and CISA guidance on AI agents that act on their own.
0 Comments