Sweden Restricts ChatGPT, Claude for Government Staff, Approves Microsoft Copilot Amid Security Concerns

Sweden’s Government Offices have restricted direct employee access to OpenAI’s ChatGPT and Anthropic’s Claude on work computers, while approving Microsoft’s Copilot following an internal security review. The move, effective October 26, 2026, aims to enhance information security within the central government administration.

This restriction means staff will now be expected to use pre-approved AI tools, with Microsoft Copilot being the sole AI assistant currently permitted for direct employee use. The decision was reported by Sveriges Radio.

Security Rationale Drives AI Tool Restrictions

Digitalisation director Mikael Wettercrantz stated that the motivation behind the restriction centers on information security, particularly concerning AI services that operate without a formal contract. “When a government employee pastes text into a consumer chatbot, the office may have little say over where that information goes next,” a concern highlighted in Sveriges Radio’s reporting, underscores the administration’s apprehension about data handling and potential information leaks.

Controlled Access to AI Models

While direct access to ChatGPT and Claude is being blocked, a limited number of staff will still be able to utilize these models through an in-house AI-assistant platform developed by the Government Offices. This approach allows the administration to maintain access to various AI models while routing them through its own controlled infrastructure, thereby managing potential security risks.

Microsoft Copilot was not automatically approved; it underwent an internal security review and required adjustments to its settings before being cleared for use. This process indicates a cautious approach to integrating AI tools into government operations.

Broader Trend in Swedish Agencies

The Government Offices are not alone in implementing such measures. The Swedish Tax Agency (Skatteverket) and the National Board of Institutional Care (SiS) have already imposed similar restrictions on AI tools. The Enforcement Authority (Kronofogden) is currently assessing its position and is expected to announce a decision soon, which could indicate a broader trend across Swedish agencies.

This latest move follows a previous restriction by the Government Offices on the Chinese AI model DeepSeek. Additionally, an audit at the Transport Administration (Trafikverket) identified problematic use of approximately 10 open AI services, providing further context for the current security-focused directives.

Implications for AI Providers

The decision positions Microsoft Copilot as the primary beneficiary, gaining exclusive status for direct employee access. For OpenAI and Anthropic, the situation is mixed; their models remain accessible via the in-house platform, suggesting the technology itself is not rejected but rather the method of access is being controlled. This strategy allows the Government Offices to retain flexibility while prioritizing security.

Uncertainties remain regarding the specific adjustments made to Copilot’s settings and the exact number of staff who will have access to ChatGPT and Claude through the in-house platform. The detailed examples of ‘problematic use’ that prompted these concerns have also not been fully elaborated.

Broader Context

Source materials place the factual news in this context: Sweden’s central government administration is pulling the plug on two of the world’s best-known chatbots.