Swedish government employees are no longer permitted to use browser-based versions of OpenAI’s ChatGPT and Anthropic’s Claude on work computers. The restriction took effect on 26 October 2026 and is intended to reduce the risk that sensitive government information could be processed or retained without adequate contractual safeguards.
The decision is a national Swedish security measure, not an EU-wide ban. It highlights growing concerns across Europe about “shadow IT” — the use of digital tools without formal approval, security assessments or agreements governing how data is handled.
Why Sweden restricted ChatGPT and Claude
The Swedish Government Offices introduced the restriction after a review of artificial intelligence tools used by staff at the Swedish Transport Administration earlier in 2026. That review identified problematic uses of AI platforms in government work.
Officials were particularly concerned that employees could enter confidential or otherwise sensitive information into public AI services without an agreement defining how the data would be protected. Depending on a provider’s terms and technical arrangements, submitted information could potentially be processed, retained or used in ways that are not suitable for government work.
Mikael Wettercrantz, director of digitalisation at the Government Offices of Sweden, said that using AI services without appropriate safeguards could have consequences for information security.
The issue is not simply whether an AI tool is useful. Public authorities must also understand:
- what information can be entered into a system;
- where data is processed and stored;
- how long information is retained;
- whether prompts or uploaded files may be used for model development; and
- which contractual protections apply if sensitive information is mishandled.
What the Swedish government’s AI rules mean
The restriction applies to browser-based versions of ChatGPT and Claude on Swedish government work computers. It does not amount to a complete ban on artificial intelligence in the public sector.
Government employees will still be able to use AI for approved tasks, but access will be subject to tighter controls. Only a limited number of staff will be allowed to continue using ChatGPT or Claude, and they must do so through a special platform for AI assistants developed by Swedish authorities.
Microsoft Copilot is being recommended as an alternative. According to the information provided by Swedish officials, Copilot has passed a security screening process and is covered by a security agreement between Microsoft and the Swedish government.
This distinction is important. Sweden is not rejecting generative AI altogether; it is favouring systems that have been assessed, contracted and configured for official use.
Which Swedish agencies are affected?
The initial review focused on the Swedish Transport Administration, but restrictions have also been reported at other public bodies.
According to Sveriges Radio, limitations on ChatGPT and Claude have been introduced at:
- the Swedish Tax Agency; and
- the National Board of Institutional Care.
The Swedish Enforcement Authority is reviewing the issue. The precise arrangements may differ between agencies, depending on their responsibilities, information systems and data-security requirements.
DeepSeek also remains unapproved
ChatGPT and Claude have joined DeepSeek on a list of AI services not approved by the Swedish government because of data-security concerns. The source material also reports that the United States announced a ban on products and services from the Chinese AI company the previous week.
The inclusion of several major AI platforms suggests that Sweden’s approach is based on governance and data protection rather than on a single company or country. The central question is whether an authority has sufficient control over the service and the information entered into it.
Why this matters beyond Sweden
The Swedish decision adds to a wider European debate about how public bodies should use generative AI. Civil servants increasingly use AI tools for drafting, translation, summarising and research, but those uses can create risks when employees upload confidential documents or personal data.
For public administrations, the practical response is likely to involve a combination of:
- approved enterprise accounts;
- data-classification rules;
- staff training;
- technical access controls;
- supplier contracts; and
- audits of how AI systems are being used.
The Swedish measure also demonstrates why organisations need clear internal policies. Blocking a website may reduce immediate exposure, but it does not by itself resolve the need for secure alternatives or guidance on acceptable use.
Does the restriction apply across the European Union?
No. This is a Swedish government decision affecting Swedish public authorities. It is not an EU regulation, European Commission decision or European Parliament measure.
Other European governments and public bodies may adopt similar policies, but any such decision would need to be confirmed separately. The approach also differs from the European Union’s broader technology framework, including rules governing artificial intelligence and data protection.
What happens next?
Swedish authorities will need to continue assessing which AI services can be used safely and for which categories of work. Agencies may also develop controlled access systems that allow staff to benefit from generative AI while limiting exposure of confidential information.
The review by the Swedish Enforcement Authority could provide another indication of how individual agencies are interpreting the national security concerns. Further guidance may also clarify which tasks are permitted, which information must never be entered into an AI service and how approved systems should be monitored.
Conclusion
Sweden’s decision to restrict browser-based ChatGPT and Claude for government staff is a targeted data-security measure, not a blanket rejection of artificial intelligence. The key takeaway is that public-sector AI use is increasingly being tied to formal contracts, security screening and controlled access. For governments across Europe, the challenge will be to preserve the productivity benefits of generative AI without exposing sensitive information to systems they cannot adequately govern.