The restrictions are intended to ensure that civil servants use AI tools that have undergone the organization's approval process. Regeringskansliet has not publicly explained the specific reasons for blocking direct access to ChatGPT and Claude, and its press office declined to disclose the security modifications made to Copilot.
Mikael Wettercrantz, the Government Offices' digitalization director, declined an interview with Ekot. However, in a message published on the organization's internal network, he warned that using AI services without established agreements could create information security risks.
The policy reflects a distinction between allowing employees to use AI and controlling how they access it. Although ChatGPT and Claude will no longer be directly available on government work computers, the internally managed platform will preserve access for certain personnel. The reporting does not specify which employees qualify for that exception or describe the platform's technical arrangements.
Microsoft Copilot will become the designated alternative for affected staff. Its approval followed a security review, although the Government Offices has not released details of the safeguards introduced. The decision therefore establishes which services employees are permitted to access without providing a public comparison of the security or performance of the competing AI systems.
Several other Swedish public agencies have already introduced similar restrictions. The Swedish Tax Agency and the National Board of Institutional Care prohibit access to comparable AI services, while the Swedish Enforcement Authority is considering implementing its own controls.
Concerns about employee use of external AI tools have also emerged elsewhere in the Swedish government. An audit conducted earlier this year at the Swedish Transport Administration found that personnel had used approximately ten publicly available AI services.
The review identified problematic practices involving sensitive information submitted to AI systems and the handling of AI-generated responses. However, the reporting did not identify the specific services or information involved, and no security breach was reported.
Sweden's public sector has simultaneously been exploring ways to incorporate AI into government operations. An October 2025 Ministry of Finance summary covering 25 agencies found that many were investigating potential applications, although adoption remained at an early stage in several organizations.
Some agencies have tested multiple commercial assistants under controlled conditions. The Swedish Post and Telecom Authority conducted trials involving both Copilot and ChatGPT with strict requirements for handling information. Kammarkollegiet also evaluated the two services before introducing digital assistants for its workforce.
These examples show that Sweden's public institutions have taken different approaches to AI deployment, ranging from restricted access to supervised testing and broader employee availability.
The Government Offices' decision also comes during a separate examination of how AI and cloud services interact with Swedish public-records law. On September 3, 2026, the Ministry of Justice appointed Lisa Englund Krafft to investigate the application of offentlighetsprincipen, Sweden's principle of public access to official documents, when authorities use AI systems and external cloud providers.
That investigation is scheduled to produce its findings by November 15, 2027. The government has not indicated that the review prompted the restrictions at Regeringskansliet.
Sweden has also outlined broader ambitions for AI adoption through a national strategy published in February 2026. The strategy seeks to strengthen the country's position in artificial intelligence while public institutions continue evaluating how the technology can be used in their operations.
Meanwhile, assessments of AI-related security risks have not been uniform across government bodies. Sweden's National Institute of Economic Research, for example, classified the information security risk associated with tools such as Copilot as very low in its own evaluation.
The Government Offices' policy does not establish that ChatGPT or Claude caused a security incident or that Copilot is inherently more secure. Instead, the reported changes center on organizational approval, information handling and the conditions under which employees may use external AI services.
When the restrictions take effect on October 26, most affected employees will be expected to use the approved Microsoft service, while selected staff can continue accessing other AI systems through the government's internal platform. The arrangement allows Regeringskansliet to maintain AI capabilities for its workforce while placing tighter controls on the services available through official computers.
This analysis is based on reporting from remio.
Image courtesy of withlocals.
This article was generated with AI assistance and reviewed for accuracy and quality.
