Sweden’s Government Offices will block access to the browser interfaces of OpenAI’s ChatGPT and Anthropic’s Claude on staff computers from Monday, October 26, 2026. Employees will instead be directed to approved AI services, including Microsoft Copilot.
The restriction also leaves another route open: the Government Offices told SVT that employees would retain access to the named language models through Intric, its internal AI platform. That detail makes this a workplace service-approval decision, not a nationwide consumer ban or a blanket prohibition on the underlying models. SVT’s report, published October 23 and updated October 24, includes the office’s statement.
The Restriction Applies to Staff Computers
The Government Offices, or Regeringskansliet, supports Sweden’s prime minister and cabinet. Its decision concerns access from the organization’s work computers.
SVT initially reported the restriction, citing Sveriges Radio’s Ekot, then added a direct response from the Government Offices specifying the browser interfaces and the October 26 start date. The office also said all employees have access to Microsoft’s AI tool.
That scope matters. A workplace block controls what staff can reach through their employer’s equipment. It is different from restricting what Swedish residents can use. The published statement does not announce restrictions on personal phones or home computers.
Other Swedish public bodies have taken similar steps. SVT reports that the Tax Agency and the National Board of Institutional Care have already introduced restrictions, while the Enforcement Authority is considering the issue. These are separate organizational decisions, not evidence of one nationwide policy.
Three AI Access Routes Receive Different Treatment
The most useful way to read the announcement is to separate the service employees open from the model that generates an answer.
| Access route | Reported treatment | Practical consequence |
|---|---|---|
| ChatGPT and Claude browser interfaces | Blocked on Government Offices staff computers from October 26 | Employees cannot continue using those interfaces from the covered computers |
| Microsoft Copilot | Remains approved following a review and increased security | Staff retain an approved assistant |
| Language models accessed through Intric | Continued access, according to the office’s statement to SVT | Blocking a chatbot interface does not necessarily remove access to its associated models |
The SVT statement is particularly important here. It identifies both the blocked browser interfaces and continued model access through Intric in the same response.
A chatbot interface and its underlying model are different parts of a service. The interface handles the user’s interaction; the model generates the response. An organization can therefore make different approval decisions about different ways of reaching a model.
My reading is that the Intric exception is the clearest evidence of the policy’s actual boundary. If the aim were simply to prohibit the named model families everywhere, retaining them through an internal platform would be inconsistent with that aim. The published decision instead distinguishes access routes.
The exact model versions, Intric architecture, employee eligibility, contracts and data-handling controls are undisclosed. Internal access should not be read as proof that models run locally, that prompts stay in Sweden, or that every employee receives identical access.
Contracts and Security Review Explain Copilot’s Approval
The stated rationale centers on information security and organizational approval.
Computer Sweden reported on October 23 that digitalization director Mikael Wettercrantz warned staff about AI services used without agreements. In an intranet comment quoted by the publication, he said such use could have “konsekvenser för vår informationssäkerhet,” meaning consequences for the organization’s information security.
Computer Sweden also reported that the Government Offices had increased Copilot’s security. SVT describes Copilot as approved after a review of the service.
Those reports explain why an approved service can receive different treatment from an unapproved one. They do not provide a comparative security test of Microsoft, OpenAI and Anthropic products.
For a workplace buyer, the consequential distinction is the actual service arrangement: what the organization has reviewed, what agreement applies and which settings it can control. A familiar product name alone cannot answer those questions.
I would treat Copilot’s approval as specific to this organization’s reviewed setup. The reporting does not identify the Copilot edition, configuration or contractual terms, so another employer cannot assume its own deployment has the same protections simply because it carries the same brand.
What Workplace AI Policies Can Learn From This
The Swedish example supports a practical approach: define approved access routes explicitly, then explain what employees may do through each one.
The reasoning follows from the published policy. ChatGPT and Claude browser access is blocked, Copilot remains available, and Intric preserves another route to models. A rule saying only “Claude is banned” would fail to describe that arrangement accurately.
For organizations designing their own policies, I’d draw four lessons:
- Name the approved interface or service. Specify whether permission covers a website, managed application or internal platform. A model name alone may be too broad.
- Define the covered equipment. State whether the rule applies to work computers, managed browsers, mobile devices or other environments. Sweden’s published statement specifically identifies staff computers.
- Document the permitted alternative. If employees should use an internal platform, explain who can access it and which tasks it supports.
- Approval should come with clear instructions about permitted inputs and how generated answers should be checked before use.
Frequently Asked Questions
3 questions
1When does the ChatGPT and Claude block begin?
The restriction takes effect on Monday, October 26, 2026, and covers browser access from Government Offices staff computers.
