The Swedish government has decided to tighten rules on the use of popular artificial intelligence platforms in public institutions. Starting October 26, Swedish civil servants are prohibited from using browser-based versions of ChatGPT and Claude on office computers. The restriction was introduced to reduce the risk of confidential information leaks and raise the overall level of data protection in the public sector.
Reason: Analysis of Practices in the Transport Agency
According to Swedish authorities, the decision was made after an analysis of artificial intelligence tool usage was conducted within the Swedish Transport Administration earlier this year. During the review, several instances of AI platform usage in the line of duty were deemed inappropriate, raising concerns about potential information security risks and triggering broader restrictions.
What Exactly is Forbidden and What is Allowed
It is important to emphasize that this is not a total ban on artificial intelligence in the workplace. Civil servants may still use AI to support their work, but this is now permitted only through regulatory-approved platforms and under stricter security measures. Direct browser-based access to ChatGPT and Claude, as was previously the case, has been closed.
Under the new rules, only a limited number of employees are allowed to continue working with ChatGPT and Claude — but exclusively through a specialized artificial intelligence platform developed by the Swedish authorities themselves. This approach makes it possible to control data flow and eliminate uncontrolled information transfer to external models.
Risk Assessment: Data May End Up in Model Training
Swedish authorities warn that transferring confidential data into artificial intelligence models without special protection and security agreements could lead to this data being stored or even used to train the models themselves. Mikael Vetterkrants, Chief Digital Officer for the Swedish government, emphasized that using technological platforms outside the jurisdiction of regulatory agencies could have serious consequences for the state's information security.
Agency Reactions and Next Steps
According to Swedish public radio Sveriges Radio, measures to restrict the use of ChatGPT and Claude have already been adopted by the Swedish Tax Agency and the National Prison and Probation Administration. Meanwhile, the Swedish Police Authority is conducting a review to consider applying similar restrictions. Thus, a practice that began with an analysis of a single agency is gradually spreading to key public administration structures.
