Starting October 26, Swedish civil servants will no longer be able to access ChatGPT or the Claude chatbot at work. The Swedish Government Offices, which serve the prime minister and ministers, have blocked the web versions of the two best-known artificial intelligence assistants on work computers. Microsoft Copilot remains the only approved tool available to officials. The government is concerned that sensitive state information could end up on servers over which it has no control.
An inspection conducted this year at the Swedish Transport Administration examined how its employees use artificial intelligence. Management was not pleased with the results. An internal audit uncovered around ten different AI services that employees were using extensively.
Director of Digitalization Mikael Wettercrantz announced the decision to colleagues on the internal network. According to him, several cases of problematic use of publicly available AI services had emerged. The issue was not only sensitive information being entered into chatbots, but also how the generated responses were subsequently handled. Wettercrantz warned that using these services without a proper contract could compromise the information security of the entire agency.
Where the information you enter into a chatbot goes
To understand what concerns the Swedish government, it is enough to consider how a typical chatbot works. When an official enters a query into a freely available version of ChatGPT or Claude, the text is sent to the provider's servers. If there is no contract between the agency and the company, there are no binding rules governing how the data will be handled. The company may store it and use it to train its model, allowing the artificial intelligence to continue learning from it.
The concerns extend beyond the technology companies themselves. US law allows government authorities there to demand data from companies. Once information is processed in a data center in the United States, it leaves Europe, where it is protected by strict EU data protection rules. Without a contract, it is also very difficult to trace who sent what to the tool and when. Cybersecurity experts refer to this phenomenon as shadow IT. It describes a situation in which employees independently obtain applications that their employer has not approved and cannot monitor.
Copilot is different. It has passed a security review and is covered by a contract between Microsoft and the Swedish government that precisely defines the rules for handling entered data. Following Copilot's review, the Government Offices further tightened its security settings. Neither ChatGPT nor Claude has undergone a similar process. However, this does not mean a complete end for the two banned chatbots. A limited group of employees may continue using them through a special platform for AI assistants developed by the Swedish authorities themselves. The press office did not specify how many people are covered by the exemption or what rules apply on the platform.
Other EU countries are also concerned
The Government Offices are far from the only Swedish institution to have curtailed the use of chatbots. According to Swedish public radio, the Swedish Tax Agency and the National Board of Institutional Care have also imposed similar restrictions on consumer AI tools. The Swedish Enforcement Authority is still preparing its own decision.
ChatGPT and Claude thus join China's DeepSeek on the list of unapproved tools. Sweden had already banned it for government work due to concerns about violations of European data protection rules. In addition, the United States recently announced a ban on all products and services from the Chinese company, fearing that local laws could give the authorities access to the company's servers.
European governments' distrust of chatbots has been growing since 2023, when Italy temporarily banned ChatGPT. The European Parliament later went one step further and completely disabled artificial intelligence features on all work devices, including Copilot. It concluded that it could not guarantee the security of the data being processed. Sweden chose a more cautious approach, approving one US company's product while blocking two others.
However, the line between a prohibited and an approved tool may not be as clear-cut as it first appears. Microsoft has integrated Anthropic's Claude language models into Copilot. They are disabled by default in the European Union, and administrators must deliberately enable them. However, the Swedish Government Offices have not yet confirmed whether they have activated them in their version. Officials could therefore theoretically continue using Claude, just under a different name.
Sources: techradar.com and euronews.com
