Sweden Blocks ChatGPT and Claude Across Government Offices

Sweden's Government Offices have implemented a ban on popular artificial intelligence services ChatGPT and Claude across all staff work computers. The restriction, first reported by public broadcaster Sveriges Radio, prevents government employees from accessing the consumer AI tools developed by OpenAI and Anthropic on state-issued devices. The measure applies to central government ministries and departments responsible for assisting cabinet ministers and shaping national policy.

Internal notices distributed across the Government Offices intranet cited significant information security concerns and the absence of binding agreements as the main drivers behind the decision. Swedish officials determined that entering public-sector data into commercial platforms poses unacceptable risks to confidentiality and document management. Without specific enterprise contracts in place, administrators concluded that external commercial services could not guarantee the required level of data control.

Rather than barring artificial intelligence tools entirely, the administration is steering civil servants toward approved, contracted enterprise software. Staff members seeking automated assistance are instructed to rely on Microsoft Copilot, which operates under established government procurement agreements and enterprise security parameters. This approach permits routine administrative use while ensuring data flows adhere to negotiated terms of service.

The policy change inside the Government Offices reflects a growing consensus across Sweden's wider public administration. Major state institutions, including the Swedish Tax Agency and the National Board of Institutional Care, previously restricted unauthorized generative AI applications from their networks. Public administrators across the country have expressed growing unease over uncontrolled data sharing on third-party cloud infrastructure.

European government bodies have increasingly scrutinized generative AI systems over how prompts, documents, and user inputs are processed and stored. Without specialized data-handling contracts, queries submitted to public-facing consumer models risk being stored on remote servers or processed under terms that conflict with European data protection frameworks. For state employees handling sensitive correspondence, such ambiguities represent a substantial regulatory vulnerability.

For high-level government offices overseeing foreign affairs, defense, and economic planning, maintaining strict custody over internal information is both a legal mandate and a national security priority. By blocking direct access to services lacking verified enterprise safeguards, Swedish IT authorities aim to prevent inadvertent leaks of state secrets and preliminary policy proposals. The ban ensures that government correspondence remains strictly within audited boundaries.

The move underscores the delicate balance modern governments must strike between harnessing technological innovation and protecting sovereign information. While tools like ChatGPT and Claude offer considerable gains in drafting texts and summarizing technical reports, uncontrolled consumer adoption complicates institutional cybersecurity. Public sector IT leaders are therefore shifting from informal employee adoption to centralized, managed rollouts.

Certain specialized departments within the Swedish state apparatus continue to test generative technology through securely managed, internal environments. These controlled frameworks enable personnel to experiment with automated workflows under isolated architectures where data remains insulated from external commercial training datasets. Such systems allow public agencies to evaluate potential productivity gains without exposing sensitive governmental files to public clouds.

As digital sovereignty concerns gather momentum across Europe, other government institutions are likely to review their own operational boundaries regarding generative AI. For Sweden's government workforce, day-to-day administrative tasks will proceed exclusively through cleared commercial channels and vetted internal tools. Technology supervisors will maintain tight oversight as public institutions navigate the evolving security landscape of advanced machine intelligence.