Sweden’s Government Offices Block ChatGPT and Claude, Keep Microsoft Copilot

The directive, which takes effect on Monday, funnels civil servants toward a single sanctioned alternative: Microsoft’s Copilot. The decision marks a significant escalation in how European public institutions manage the delicate balance between embracing technological efficiency and protecting sensitive state information from foreign or unauthorized exposure.

The rationale behind the sweeping block was detailed by Mikael Wettercrantz, the Government Offices’ director of digitalisation, in an internal memo posted to the staff intranet. Wettercrantz underscored that utilizing generative artificial intelligence services without a formal, legally binding enterprise contract carries profound and potentially dangerous consequences for public sector information security.

“We have seen several examples of problematic use of open AI services,” Wettercrantz explained in the internal communication. “The examples concern both the sensitivity of the information that has been entered and how the AI-generated responses have been handled.”

At the heart of the restriction lies a fundamental architectural difference between consumer-grade platforms and enterprise-level software. Standard consumer AI tools operate without bespoke government security agreements, meaning that user queries, drafts, and administrative notes are routinely transmitted to external company servers. Under foreign jurisdictions—particularly in the United States, where many of these foundational AI developers are based—local laws can compel technology corporations to surrender stored data to domestic authorities upon request, bypassing traditional international judicial cooperation frameworks.

In contrast, a formal enterprise security agreement establishes legally binding boundaries. It explicitly dictates what a service provider can and cannot do with input data submitted by government employees, ensuring that sensitive state information is not harvested for model training or exposed to third-party oversight. This distinction between a vetted enterprise contract and an open consumer tool forms the bedrock of Sweden’s new administrative policy.

Microsoft’s Copilot cleared this rigorous internal hurdle after the Government Offices conducted a thorough security review and subsequently elevated its default security and privacy settings. Notably, a comparable vetting process was not carried out for ChatGPT or Claude. Despite the broad ban on desktop applications, some government employees will retain residual access to the blocked tools through an internal AI platform developed specifically by the Government Offices, though the press office has declined to specify the exact reach of this platform or the precise restrictions governing its use.

Sweden’s central administration is far from alone in adopting a cautious stance toward consumer AI. Several other Swedish public agencies have already instituted similar bans or are actively moving to lock down their digital environments. The Swedish Tax Agency and the National Board of Institutional Care have previously implemented restrictions on consumer-grade artificial intelligence tools. Meanwhile, the Enforcement Authority, the government agency responsible for debt collection and enforcing court orders, is currently preparing its own restrictive policy.

Internal audits across other sectors of the Swedish civil service have revealed a pressing need for oversight. A recent internal audit conducted by the Swedish Transport Administration uncovered approximately ten different artificial intelligence services being utilized across the agency in what investigators characterized as a "widespread and risky way."

These localized moves in Sweden mirror a broader continental anxiety regarding data sovereignty and artificial intelligence. In February, the European Parliament took an even more drastic approach by completely disabling AI features across all professional work devices—sweeping ChatGPT, Claude, and Copilot into a total blackout after concluding that it could not reliably guarantee the security of the data processed by those systems.

European government concerns regarding artificial intelligence data security have steadily mounted since Italy’s prominent temporary ban on ChatGPT in the spring of 2023. However, Sweden’s recent policy carves out a distinct path compared to the blanket prohibitions favored by institutions in Brussels. Rather than shutting the door on generative technology altogether, Stockholm is drawing a narrower line, sanctioning one major U.S. technology product while barring two of its primary competitors.

Yet, the technological reality underlying these corporate platforms complicates the government’s clean bifurcation between "safe" and "unsafe" systems. In early 2026, Microsoft integrated Anthropic’s Claude models directly into its broader Copilot ecosystem. Within the European Union, however, Anthropic’s models are configured by default to remain turned off, requiring system administrators to deliberately and manually enable them.

Sweden’s Government Offices have not publicly confirmed whether their specific corporate instance of Copilot has Claude switched on under the hood. As a result, the practical distinction between the officially banned AI services and the officially approved platform may be considerably less absolute than the government’s administrative announcement implies, highlighting the complex and fast-moving challenges facing public sector institutions attempting to regulate a rapidly evolving technological landscape.