Sweden bans ChatGPT and Claude on government computers

The new government ban

The Swedish government has officially blocked access to ChatGPT and Claude on all staff computers as of this Monday. Civil servants are now required to use Microsoft Copilot as their sole authorized tool for artificial intelligence tasks. This decision marks a significant shift in how the state manages digital tools for its employees.

Mikael Wettercrantz, the director of digitalization for the Government Offices, communicated the change to staff through the internal network. He explained that the move is necessary to protect sensitive government information from various security risks. The government is prioritizing the safety of national data over the convenience of using multiple different platforms.

This action follows a period of observation where officials noticed employees using various AI services in ways that could compromise security. The government has now moved to centralize these tools under a single, vetted provider. By restricting access, the state aims to create a controlled environment for all AI-assisted work.

Security risks of consumer AI

The primary concern for Swedish officials involves the way consumer AI tools handle user input. Many of these services may send questions and other data to servers operated by the companies providing the software. This creates a significant vulnerability when government employees enter confidential or private details into these systems.

Furthermore, legal frameworks in the United States can sometimes allow authorities to demand data from private companies under specific circumstances. Without formal agreements in place, there is little protection for the information shared with these providers. The government argues that using AI tools without proper contracts can lead to unintended information leaks.

The Swedish government has already reviewed Microsoft Copilot and adjusted its security settings specifically for government use. In contrast, the government did not perform the same approval process for ChatGPT or Claude. This distinction suggests that the approved tool has specific protections that the others currently lack in a public setting.

Impact on government staff

The ban affects a wide range of civil servants who may have previously relied on various AI tools for their daily tasks. While the main platforms are blocked, some employees might still have access to an internal AI platform developed by the Government Offices. However, the government has not yet explained how many people can use this internal system or what specific restrictions apply to it.

The move is part of a broader trend across different Swedish agencies. The Tax Agency and the National Board of Institutional Care have already introduced similar restrictions on consumer AI tools. The Enforcement Authority is also currently preparing its own decision regarding how it will handle AI technology.

An internal audit at the Transport Administration recently found that around 10 different AI services were being used in a widespread and risky manner. These findings likely served as a catalyst for the current ban. The government is now working to ensure that all departments follow a unified and secure standard for AI usage.

Technical nuances and uncertainty

There is some lingering uncertainty regarding the specific technology that will be available to Swedish workers. Microsoft recently added Anthropic’s Claude models to its Copilot service, but this is scheduled for early 2026. In the European Union, these models are turned off by default and require administrators to manually enable them.

The Government Offices have not confirmed whether they have enabled the Claude models within their version of Copilot. This leaves a question about whether the approved tool will eventually include the same capabilities as the blocked ones. For now, the government is sticking to the settings they have already vetted and secured.

The decision reflects a cautious approach to the integration of large language models in public administration. While the government wants to utilize AI, it is doing so only when it can guarantee the privacy of the data. This creates a tiered system where only specific, contractually protected versions of AI are permitted for official business.

The broader European context

Sweden's actions are not unique, as other bodies in Europe have taken similar steps to manage AI risks. The European Parliament adopted a broader approach in February by disabling AI features across all its work devices. This included a ban on ChatGPT, Claude, and even Microsoft Copilot because they could not guarantee data security.

The Parliament's decision highlights a fundamental disagreement over whether current AI infrastructure can be trusted with sensitive information. While some organizations seek to find a middle ground, others believe that the risks of data leakage outweigh the benefits of the technology. Sweden's choice to allow Copilot suggests they believe a managed version of the technology is viable.

As more countries grapple with these issues, the distinction between consumer and enterprise AI will become increasingly important. Governments are looking for ways to harness the power of AI while maintaining strict boundaries on who can see the data. Sweden's policy serves as a current example of how these security concerns are being translated into official government policy.

What to watch next

Observers will be watching to see how the Government Offices configures its internal AI platform and what specific restrictions will be placed on it. It will also be important to see if they eventually enable Claude models within their Copilot environment as they become available.

Additionally, the final decisions from the Enforcement Authority and other agencies will provide a clearer picture of the national standard. The government may also need to provide more clarity on the specific 'problematic use' cases that led to this ban to help employees understand the boundaries of acceptable AI use.

Try it

Check the official website of the Swedish Government Offices to see if they release any public documentation on their approved AI security standards.