Stockholm, 26 October. Since 26 October, staff of the Swedish Government Offices have been unable to open the browser versions of OpenAI's ChatGPT and Anthropic's Claude on their work computers, the Government Offices themselves told SVT in an email.
Swedish public radio Sveriges Radio was first to report the decision on 23 October, having obtained an internal staff notice.
The Government Offices are the administrative machinery serving the prime minister and the ministries. Their digitalisation director, Mikael Wettercrantz, set out the reason in the internal notice: using AI services without an agreement with their owner can have consequences for information security; according to Sveriges Radio, the offices had also seen several cases of problematic use of such services. Wettercrantz declined an interview with the radio.
Staff have been advised to use Microsoft Copilot instead: it has passed a security review and is covered by a data-protection agreement between Microsoft and the Swedish authorities. The language models themselves are not banned: "Within Intric, staff still have access to the ChatGPT and Claude language models," the Government Offices wrote to SVT; Intric is a Swedish platform for AI assistants through which access is provided under contract. The Chinese service DeepSeek is also among the services not approved for direct use in the government offices.
According to the Swedish public broadcaster Sveriges Radio, similar restrictions already apply at the Tax Agency and the National Board of Institutional Care, and the Enforcement Authority is considering similar measures.
What it means: Sweden is not banning AI in government, but permitting only services whose owners have signed a data-protection agreement; others are available only through its own secure platform.
Who benefits: Microsoft, whose Copilot has become the Swedish government's recommended tool; citizens, who face a lower risk of their data leaking through chatbots. OpenAI and Anthropic lose direct access to some government users until they sign such agreements.
Background: Kazakhstan has taken a similar approach: civil servants may not use AI models that process data on foreign servers, and official correspondence and documents are processed on domestic computing capacity, First Vice Minister of Artificial Intelligence and Digital Development Rostislav Konyashkin has said.
