The administrative engine of the Swedish government, the Government Offices, has announced a significant shift in its digital policy, moving to restrict access to major artificial intelligence platforms. Starting this Monday, employees within the body that supports the prime minister and cabinet will be barred from using ChatGPT and Claude on official computers. Under the new directive, Microsoft’s Copilot has been designated as the sole authorized AI tool for civil servants.

The decision, first reported by Sweden’s national, publicly funded broadcaster Sveriges Radio, marks a tightening of the government’s stance on how emerging technologies are integrated into the public sector. The rationale behind this move centers on the potential for sensitive information to be mishandled by consumer-grade AI services that lack the stringent data protection frameworks required by state entities.

Mikael Wettercrantz, the director of digitalisation for the Government Offices, communicated the change via the staff intranet, framing the decision as a necessary precaution to protect national information security. According to Wettercrantz, the usage of AI services in the absence of a formal, vetted contract creates unacceptable risks for the government.

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

The Risks of Consumer-Grade AI in Government

At the heart of the government’s concern is the fundamental architecture of public-facing AI tools. Consumer versions of chatbots like ChatGPT and Claude typically operate by sending user queries to external company servers. In many cases, these inputs are processed to train future iterations of the underlying models. While convenient for casual users, this creates a major liability for government institutions handling classified or confidential data.

Furthermore, these tools are subject to the legal jurisdictions of their providers. Because the primary developers of these models are based in the United States, their operations are governed by U.S. law, which can, under certain circumstances, allow authorities to compel these companies to disclose user data. Without a formal, enterprise-grade security agreement, the Swedish government cannot guarantee that its data remains sovereign or protected from external legal requests.

A formal security agreement is essential because it defines the boundaries of data usage, explicitly stating what a provider can and cannot do with government inputs. By transitioning to a vetted enterprise contract—such as the one established for Microsoft’s Copilot—the government can enforce specific security settings, ensure data encryption, and maintain control over how information is processed and stored.

The Government Offices confirmed that Copilot was only approved after a rigorous security review, which included the implementation of enhanced safety settings that were not—and could not be—applied to the open versions of ChatGPT or Claude. While the government has built an internal platform that allows some employees to access the restricted tools under specific, controlled circumstances, the press office has not provided details regarding the scale of this access or the exact security protocols in place for that platform.

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

A Growing Trend Across Swedish Agencies

The move by the Government Offices is not an isolated incident but rather part of a broader trend of caution spreading across Swedish public agencies. The Swedish Tax Agency and the National Board of Institutional Care have already implemented similar restrictions, banning the use of unvetted, consumer-facing AI tools on their networks.

The Enforcement Authority, the government agency tasked with debt collection and the execution of court orders, is currently in the process of finalizing its own decision regarding the use of such technology. The climate of caution is further reinforced by internal audits, such as the one conducted by the Transport Administration, which reportedly uncovered the widespread and, in the auditor’s view, risky use of roughly ten different AI services across the agency. These findings have served as a wake-up call for state departments, prompting a shift toward centralized, authorized AI solutions.

European Context and the Question of Consistency

Sweden’s actions reflect a larger, continent-wide struggle to balance the efficiency of AI with the imperative of data sovereignty. The European Parliament took an even more aggressive stance in February, opting to disable all AI features across its work devices—including ChatGPT, Claude, and Copilot—after concluding that it could not sufficiently guarantee the security of the data processed by these models.

The unease among European governments regarding data security has been building steadily since Italy’s landmark decision to temporarily ban ChatGPT in 2023. Since then, national regulators and government IT departments have been tasked with defining the "red lines" for AI deployment. Sweden’s current policy creates a specific, if nuanced, distinction: by authorizing one U.S.-based product while excluding others, the government is attempting to strike a balance between innovation and risk management.

However, the distinction between "approved" and "banned" tools may be more complex than the policy suggests. Since early 2026, Microsoft has integrated Anthropic’s Claude models directly into its Copilot ecosystem. Within the European Union, however, these models are typically set to "off" by default, requiring administrators to manually and deliberately enable them.

The Government Offices has not confirmed whether its specific instance of Copilot has the Claude models activated, leading to questions about whether the official ban on Claude is fully realized in practice. If the underlying technology is accessible through the "approved" gateway of Copilot, the security implications may be more uniform than the official rhetoric suggests.

As the Swedish government continues to navigate the complexities of digital governance, the policy highlights the difficult reality of the current AI landscape: technology that promises to revolutionize productivity also poses existential questions about the security of the information upon which the state relies. For now, the directive remains a clear message to civil servants that while AI is part of the future, its adoption will be strictly managed under the watchful eye of the government’s information security experts.

Leave a Reply

Your email address will not be published. Required fields are marked *