Sweden Blocks ChatGPT and Claude on Government Computers

Sweden blocks ChatGPT and Claude on staff work computers at the Government Offices from Monday, directing employees to approved AI tools instead. The restriction applies to the Regeringskansliet, the administrative body that supports the prime minister and the cabinet, and it was first reported by Sveriges Radio's Ekot news desk.

Microsoft's Copilot is the one major chatbot left standing. It stays available to all staff after passing a security review that led the offices to raise its security settings. A limited number of employees will still reach the ChatGPT and Claude models through Intric, an internal AI assistant platform the offices built in-house.

Why Stockholm is drawing the line

Ekot first reported the restriction, citing internal information it had seen. The Government Offices then confirmed the change in an email to broadcaster SVT, saying the browser interfaces for the two services would no longer be reachable from work computers from Monday.

Ekot published its report on October 23, citing internal documents and information from inside the offices. Wettercrantz's intranet note did not point to a leak or a single incident, and none of the reporting says a breach prompted the decision. The stated logic is the absence of a contract, not a documented failure.

The stated reason is information security, not a single incident. Digitalisation director Mikael Wettercrantz wrote on the offices' intranet that using AI services without a formal agreement can affect information security. He declined an interview with Ekot, and the press office declined to detail the Copilot changes, citing security.

According to internal information seen by Ekot, auditors found several examples of problematic use of open AI services. The examples involved both the sensitivity of information staff entered into the tools and how employees handled the answers the tools produced. Without a contract, queries sent to a consumer chatbot go to an outside company's servers, and American law can require companies to hand over data.

Part of a wider Swedish clampdown

Copilot did not get approved automatically. Internal documents show the offices raised its security settings after a review, and the press office refused to specify the changes, citing security. The move itself is an internal IT decision, not legislation, and staff who need the blocked models for specific work can still reach them through the internal platform, although the press office has not said how many employees qualify or under what conditions.

The Government Offices are not acting alone. The Swedish Tax Agency and the National Board of Institutional Care have introduced similar restrictions, and the Swedish Enforcement Authority is weighing a decision, while the Swedish Police Authority is expected to decide soon. An internal audit at the Swedish Transport Administration found staff using about ten open AI services widely and at high risk.

The offices had already restricted DeepSeek, the Chinese AI service, in an earlier decision. The pattern fits a broader trend beyond Sweden: Italy blocked ChatGPT in 2023, and DeepSeek has been withdrawn from United States federal civilian agencies, as governments weigh how much control they have over where official queries go.

For now, the message from Stockholm is plain. Sweden blocks ChatGPT and Claude from the open web at work, while keeping one approved assistant and one internal platform for staff who need more. Whether that balance holds will depend on how well Copilot covers the daily work of the people who used to open a chatbot tab without thinking twice.

The internal platform comes from a Swedish AI company called Intric, which won the contract to supply it this summer, according to Techtidningen. That arrangement is what lets some staff keep working with the blocked models inside a controlled environment, rather than through the public websites. For most employees, the practical change is simpler: the two public chatbot websites will no longer load on office machines, and Copilot becomes the default assistant.

The distinction the offices draw is contractual. Copilot runs under an agreement that sets out how data is handled and who is responsible for it, while the consumer versions of ChatGPT and Claude do not carry the same arrangement for government work. A contract, in this view, is what turns an outside chatbot into a tool the government can answer for.

Sources: Gobi Daily and Ranzware covered the Ekot report on the block.