Sweden's central government administration is restricting access to two of the world's best-known chatbots. Starting Monday, staff at the Government Offices will no longer be able to use OpenAI's ChatGPT or Anthropic's Claude on their work computers.

Microsoft's Copilot is the one AI assistant left standing. For the civil servants who support the prime minister and cabinet, the menu of approved chatbots has become shorter.

What the Government Offices decided

The restriction was reported by Sveriges Radio, Sweden's national, publicly funded radio broadcaster. It applies to the Government Offices, known in Swedish as Regeringskansliet, the administrative machinery behind the cabinet. According to the reporting, the block takes effect on October 26, 2026. From that date onward, employees are expected to stick to pre-approved AI tools — in practice, that means Copilot.

Copilot did not receive its approval automatically. It passed an internal security review, but only after adjustments were made to its settings. In other words, the deciding factor was not what the tools can do, but under what terms they operate.

ChatGPT and Claude are not disappearing from the building entirely. A limited number of staff will still be able to reach both models through an AI-assistant platform the Government Offices built in-house.

The motivation, as outlined by digitalisation director Mikael Wettercrantz, centers on information security, with the concern especially sharp around AI services that operate without a contract in place.

“We have seen several examples of problematic use of open AI services,” according to Sveriges Radio's reporting.

The worry is straightforward: when a government employee pastes text into a consumer chatbot, the office may have little say over where that information goes next. For an administration at the center of government, that loss of visibility is exactly what the pre-approval system is designed to close.

A pattern across Swedish agencies

The Government Offices is not moving alone. The Swedish Tax Agency, Skatteverket, has already imposed similar restrictions on AI tools, and so has the National Board of Institutional Care, known as SiS. The Enforcement Authority, Kronofogden, is assessing its own position and is expected to announce a decision soon. Read together, the moves already taken point in one direction: across Swedish public administration, AI access is being routed through reviewed and approved channels rather than left open to individual staff.

There is precedent inside the Government Offices as well. The administration had already restricted DeepSeek, the Chinese AI model, before latest move.

An audit at the Transport Administration, Trafikverket, offers a glimpse of why officials are nervous: it identified problematic use of around 10 open AI services. It is this kind of documented misuse that the tightening of access is intended to head off.

What this means for AI vendors and public-sector users

Microsoft's position changes the most, as Copilot is now the only approved AI tool for staff at the Government Offices. For OpenAI and Anthropic, the signal is mixed. Their models remain in use through the in-house platform, so the technology itself has not been rejected. What has been rejected is the open, direct-to-employee route. Staff can no longer simply open a browser tab and start typing; access now runs through a channel the government controls.

The in-house platform approach is notable. It allows the Government Offices to keep access to multiple models while routing them through infrastructure it manages. For vendors selling into the public sector, the case underlines that approval can hinge as much on contractual coverage and security settings as on what a model can do.

The next development to watch is Kronofogden, whose expected decision will show whether the restrictive approach continues to spread through Swedish agencies.