Sweden blocks ChatGPT and Claude for civil servants

The use of generative AI services in public administration has grown faster in recent years than organisations' own rules for them. When employees paste sensitive text into free versions of chatbots such as ChatGPT (from OpenAI) and Claude (from Anthropic) without a data processing agreement, there is no guarantee the information won't be stored by the provider or used to further train the models — something governments and larger organisations increasingly call shadow AI. The case fits a broader pattern of European authorities tightening which AI tools staff may use at work.

The internal audit that set the process in motion

The ban follows a review the Swedish Transport Administration carried out earlier this year, which found widespread and risky use of around ten different AI services among its staff. In its report, the internal audit wrote: "We have seen several examples of problematic use of open AI services. The examples concern both the sensitivity of the information entered and how the AI-generated answers were handled."

The findings prompted the Government Offices — the body that supports the prime minister and the ministries — to tighten its own rules. As of 26 October, staff have no longer been able to reach ChatGPT or Claude through a browser on their work computers.

Copilot made safer after review

The goal is for staff to use services the Government Offices has approved, chiefly Microsoft's AI tool Copilot. According to internal information obtained by Sveriges Radio's news desk Ekot, the agency has also tightened security in Copilot following its own review, though the press office declined to go into detail.

– Using AI services without an agreement can have consequences for our information security, says Mikael Wettercrantz, director of digitalisation at the Government Offices, in a comment on the agency's own intranet. He declined to be interviewed by Ekot.

A limited number of staff still have access to ChatGPT and Claude, through a dedicated platform for AI assistants the Government Offices has developed itself. According to The Next Web, the platform is run by the Stockholm-based company Intric, which won a deal with the Government Offices in July and says the service runs on servers in Sweden.

Other Swedish agencies follow the same line

Ekot's survey shows the Government Offices is not alone. Both the Swedish Tax Agency and the National Board of Institutional Care have introduced similar blocks on unregulated AI use, while the Swedish Enforcement Authority is currently reviewing the issue ahead of a coming decision. The Government Offices had already restricted use of the Chinese AI service DeepSeek.

Data can cross the EU's digital borders

TechRadar points to an underlying concern beyond the agreement itself: without an enterprise agreement, information entered into an AI chatbot can be processed in data centres outside Europe, including in the US, moving beyond the protections of European rules. The absence of such an agreement also makes it harder for an organisation to audit how its own staff actually use the tools.

The same week, US authorities announced a ban on DeepSeek across all civilian federal agencies, citing similar data-security concerns.

Sources: Sveriges Radio/Ekot, Euronews, TechRadar, The Next Web

‍