The Swedish government has restricted access to OpenAI's ChatGPT and Anthropic's Claude, citing serious concerns about the protection of sensitive information. The decision took effect on Monday 26 October 2026 and applies to official computers in government offices and ministries.
According to Swedish public broadcaster Sveriges Radio, staff in the Government Offices can no longer use the two applications' public web versions directly from their work computers.
The decision is neither a general ban on artificial intelligence nor a restriction on citizens' access to ChatGPT and Claude. It specifically concerns the use of unapproved digital tools in the government workplace.
What prompted the decision
The tighter measures followed an internal audit at the Swedish Transport Administration, Trafikverket, earlier in 2026.
The audit identified problematic uses of AI applications by employees, both in terms of the sensitivity of information entered into the platforms and the way their generated answers were used.
The main concern is that official documents, personal data or other confidential information could reach external IT systems without the necessary contractual guarantees governing their handling and protection.
Mikael Wettercrantz, Director of Digitalisation at Sweden's Government Offices, warned that using AI services without approved agreements could have implications for information security.
In IT, this is known as shadow IT: employees using technological applications without approval or supervision from the responsible department.
Why Microsoft Copilot remains available
Despite the restrictions on ChatGPT and Claude, employees can still access Microsoft Copilot, which has been approved by the Swedish government services.
According to Sveriges Radio, the tool was assessed and its security measures strengthened before approval. Swedish authorities have not disclosed details of the changes, citing security reasons.
The distinction does not mean that Copilot is automatically considered safer than competing models. The key criterion is that this particular solution has been reviewed and approved for use in the government environment.
Access to ChatGPT and Claude models under controlled conditions
Swedish authorities are not abandoning OpenAI's and Anthropic's language models.
As the Government Offices confirmed to Swedish television network SVT, employees can still access the models through the specialist AI platform Intric, rather than using the public web applications directly.
The government is seeking to retain the benefits of AI while limiting the risks associated with the uncontrolled handling of official information.
Restrictions in other public services
Similar measures have already been introduced at the Swedish Tax Agency and the National Board of Institutional Care.
The Swedish Enforcement Authority is considering comparable restrictions, while use of the Chinese application DeepSeek has also been restricted within the Government Offices.
Data protection has become a central issue for the country's public administration. Sweden's privacy protection authority, IMY, has included public-sector use of AI among its supervisory priorities for 2026.
AI in public administration
Sweden's decision highlights a question facing a growing number of public services: how to use AI tools without putting personal data, confidential documents and official information at risk.
These applications can help with drafting documents, analysing data and employees' everyday work. They also require clear rules about which information may be entered and who is responsible for processing it.
The Swedish government has therefore chosen to restrict direct access to unapproved services while allowing the same technologies to be used through controlled systems.
