Sweden to Block ChatGPT and Claude on Government Computers Tomorrow — Here's Why Copilot Is Still Allowed
Sweden is preparing to block direct access to ChatGPT and Claude on government employees' work computers from October 26, in a move that puts information security and the use of artificial intelligence inside public institutions under the spotlight.
The restriction will apply to staff at the Swedish Government Offices, the administrative organisation that supports the prime minister and the country's ministries. Instead of accessing the two AI services directly through their websites, employees are being directed to Microsoft Copilot, which has been approved for official use following a security review.
The decision does not amount to a complete ban on the underlying AI models. Swedish authorities have indicated that some employees will continue to access ChatGPT and Claude through a separate internal AI platform developed by the Government Offices.
The distinction is important. The issue is not simply whether government employees can use artificial intelligence. It is whether they can use services that have been assessed and approved for handling information in an official working environment.
According to Swedish public broadcaster Sveriges Radio and news outlet SVT, the government's digitalisation director has warned that using AI services without a formal agreement could have consequences for information security. The Government Offices has also tightened security settings for its approved Copilot service after reviewing the tool.
The move follows similar restrictions at other Swedish public agencies, including the Swedish Tax Agency. It also comes amid wider concerns about employees entering sensitive information into public AI services and relying on AI-generated answers without adequate safeguards.
What Sweden Is Blocking From October 26
The restriction is scheduled to take effect on Monday, October 26, 2026. From that date, employees at the Swedish Government Offices will no longer be able to access the browser interfaces of ChatGPT and Claude from their work computers.
The decision was reported by Sveriges Radio's Ekot and subsequently confirmed to SVT by the Government Offices.
For employees, the practical change is straightforward: they are expected to use AI tools that have been approved for their official work rather than opening the ChatGPT or Claude websites directly.
Microsoft Copilot is the approved alternative identified in the reporting. The Government Offices reviewed the service and subsequently increased its security settings, although it has not publicly disclosed the technical details of those changes.
The restriction should not be interpreted as a nationwide ban on ChatGPT or Claude for ordinary Swedish citizens, private businesses or students. The reported measure concerns access from government work computers within the Government Offices.
That difference matters because a government can restrict the tools its employees use on official devices without prohibiting the same services throughout the country.
What is confirmed: the Government Offices is restricting direct browser access to ChatGPT and Claude on staff computers, while directing employees to an approved AI service.
What is not confirmed: a nationwide prohibition on the two services or a complete removal of their underlying models from every government system.
Why Sweden Is Concerned About Public AI Tools
The central concern is information security.
Government employees may work with sensitive correspondence, policy documents, internal assessments, personal information and material that should not be shared outside authorised systems. If employees enter such material into an AI service without an appropriate agreement or clear organisational safeguards, the institution may lose control over how that information is handled.
Mikael Wettercrantz, the Government Offices' director of digitalisation, said in an internal comment reported by Sveriges Radio that using AI services without a contract could have consequences for information security.
The concern extends beyond the information entered into a chatbot. Employees must also assess how they use the answers that AI systems produce, whether the responses are reliable and whether generated material is suitable for official decisions.
An AI assistant can produce a convincing answer that contains mistakes, incomplete context or unsupported claims. In a government setting, those errors can have consequences if they enter a report, recommendation or administrative decision without proper review.
The Government Offices' decision reflects an effort to establish clearer boundaries around which tools staff can use and under what conditions.
It does not, by itself, prove that ChatGPT or Claude is inherently unsafe for every use. Enterprise agreements, data-handling settings, access controls and organisational policies can differ significantly between deployments.
The relevant question for a public institution is whether a particular service has been approved for its intended purpose and whether the safeguards meet the organisation's requirements.
Why Microsoft Copilot Is Still Allowed
Sweden's approach is not a rejection of all artificial intelligence. Instead, it places greater emphasis on using an approved service.
Microsoft Copilot remains available to Government Offices staff after a review of the tool and an increase in its security settings. Officials have not publicly detailed exactly which technical controls were changed.
A formal agreement with a provider can establish requirements around data processing, access, retention, administration and other aspects of service use. Depending on the product and configuration, an organisation may also have controls that are not available when employees independently use consumer-facing services.
However, approval should not be confused with an unconditional guarantee of security. An approved AI tool still needs appropriate configuration, employee training, access restrictions and clear rules governing the information that can be submitted.
The decision also does not establish that Copilot is universally safer than ChatGPT or Claude in every environment. Different versions of these products can have different security features and contractual arrangements.
The distinction reported in Sweden is that the Government Offices has reviewed and approved Copilot for its official environment, while direct access to the other two services is being restricted.
For organisations choosing AI tools, that is a reminder that procurement and data governance can matter as much as the model's capabilities.
ChatGPT and Claude Will Not Disappear From Every Government Workflow
One detail makes Sweden's decision more nuanced than a straightforward ban.
According to the Swedish reporting, some Government Offices employees will retain access to ChatGPT and Claude through a separate internal AI platform developed by the organisation itself.
This means the restrictions focus on direct access to the public-facing services from work computers, rather than necessarily eliminating access to the underlying models in every approved workflow.
The distinction between a model and the service used to access it is important. An organisation may be able to make a model available through a controlled platform with its own authentication, administrative policies and technical safeguards.
The public reporting does not establish exactly how many employees will retain access through the internal platform or the full range of controls governing that access.
It would therefore be inaccurate to describe the measure as a complete ban on every version of ChatGPT and Claude across the Swedish government.
Instead, the policy appears designed to direct staff away from unapproved direct access and towards AI services and access routes that the organisation has reviewed.
Sweden's Tax Agency Has Already Taken Similar Action
The Government Offices is not the only Swedish public institution taking a more restrictive approach to AI tools.
Sveriges Radio has reported that the Swedish Tax Agency, known as Skatteverket, and the National Board of Institutional Care, or SiS, have also introduced restrictions on certain AI services.
The Swedish Enforcement Authority, Kronofogden, has been considering the issue, according to the same reporting. Its consideration should not be described as an implemented ban unless the agency confirms a final decision.
The developments suggest that public institutions are examining how employees use AI tools, particularly where work may involve confidential information or material subject to specific handling requirements.
Earlier concerns were also reported at the Swedish Transport Administration, Trafikverket. An internal audit identified widespread and potentially risky use of several AI services, including concerns about the sensitivity of information entered into open tools and the handling of AI-generated responses.
These cases point to a broader governance challenge. Employees may adopt AI tools faster than their organisations can establish clear policies, approved platforms and practical training.
Blocking selected websites is one way to enforce a policy, but it does not replace the need for staff to understand which information can be shared, how AI output should be checked and which services are authorised for particular tasks.
What This Means for OpenAI, Anthropic and Microsoft
The decision puts three major AI providers into the same public-sector discussion, but it should not be interpreted as evidence that Sweden has declared one company's models superior to the others.
OpenAI develops ChatGPT, Anthropic develops Claude, and Microsoft offers Copilot. Their products have different features, subscription options, integrations and enterprise arrangements. The exact safeguards available can depend on the product version and how an organisation deploys it.
For OpenAI and Anthropic, the Swedish restriction highlights a practical challenge: public institutions may limit direct access to consumer-facing services even when employees find those services useful.
For Microsoft, the decision gives Copilot a more prominent role in the Government Offices' approved AI workflow. However, the reporting does not establish that Microsoft has received an exclusive, long-term government contract covering every possible AI use.
The broader commercial lesson is that selling AI to governments and regulated organisations requires more than strong model performance. Providers must also address data protection, security reviews, contractual requirements, administrative controls and the needs of institutional customers.
For public-sector buyers, the decision underlines the importance of evaluating a service in the context in which it will actually be used. A tool suitable for drafting a public-facing summary may not automatically be suitable for processing confidential material.
Could Other European Governments Follow Sweden?
Sweden's move adds to a wider debate over how public institutions should use commercial AI systems.
Government agencies across Europe and elsewhere face similar questions: Can employees use publicly accessible chatbots for routine work? What information should never be entered into an external service? Which AI products have been reviewed? How should employees verify generated content?
There is no single answer that applies to every organisation. A government office handling classified material may need stricter controls than a department using AI to draft public communications. Legal requirements, procurement rules, contractual protections and technical configurations also differ.
Some organisations may respond by blocking selected websites. Others may provide centrally managed AI services, enterprise accounts or internal platforms that offer greater administrative control.
Sweden's decision demonstrates one possible approach: restrict direct access to selected services while continuing to support AI use through approved alternatives.
Whether other governments adopt similar policies will depend on their own security assessments and procurement decisions. The Swedish announcement alone does not establish that a coordinated European-wide ban is underway.
What Happens Next?
The restriction is scheduled to begin on October 26. The key questions are how it will be implemented, how staff will adapt to the approved tools and whether additional agencies announce similar measures.
Further details about Copilot's security configuration and the internal platform that still provides access to ChatGPT and Claude would help clarify how the Government Offices is balancing AI capabilities with information security.
Another issue to watch is whether the Swedish Enforcement Authority reaches a final decision on its own restrictions.
For now, the clearest conclusion is that Sweden is tightening control over which AI services government employees can access directly on their work computers. The measure is limited to the reported government environment, and some employees will retain access to ChatGPT and Claude through an internal platform.
The Bottom Line
Sweden is blocking direct access to ChatGPT and Claude on Government Offices work computers from October 26, 2026, while directing staff to Microsoft Copilot, which was approved following a security review.
The move follows concerns about using AI services without appropriate agreements and the risk of exposing sensitive information. Similar restrictions have been reported at other Swedish public agencies.
But the decision is not a nationwide ban on ChatGPT or Claude, and it does not eliminate access to the models for every government employee. Some staff will continue to use them through a separate internal AI platform.
The larger issue is how governments can benefit from artificial intelligence without losing control over sensitive information. As public institutions expand their use of AI, approval processes, contractual safeguards and employee training are likely to remain central to that debate.
This article is for informational purposes only.
Sources
