Sweden’s Government Offices restrict ChatGPT access
A workplace block takes effect today. Access to models through approved tools remains, making the service and its data rules the important distinction.
Sweden’s Government Offices are restricting access to the ChatGPT and Claude websites on staff computers from today, October 26. The scope matters: this is a workplace access policy, rather than a countrywide ban on using AI.
In an email published by SVT, the offices said the browser interfaces would become inaccessible on work computers from that date. Employees would still have access to the models through Intric, an approved platform. SVT first published its report on October 23 and updated it on October 24.
The service matters as well as the model
Ekot’s original reporting says staff are being directed toward approved services, including Microsoft Copilot. An internal explanation warned that using AI services without agreements could affect information security.
Ekot also reports that Copilot’s security was tightened after a review. The press office declined to provide the details. That leaves an important limit on what outsiders can conclude: the reporting does not establish which particular controls changed, or prove that every approved workflow is safe.
The practical distinction is between access arrangements. A familiar model name alone does not tell a worker which organization handles the input, what information is permitted, or which account and settings their employer has approved.
A useful check before pasting
Sweden’s digital-government agency, Digg, already recommends management oversight of public and integrated generative AI. Its January 2025 guidance says staff should understand permitted uses and information, while integrations with organizational data require attention to information management and permissions. That is background guidance, not a description of the new block’s technical implementation.
For anyone using AI at work, our suggested next step is modest: ask for a concrete answer to three questions before sharing internal material.
- Which service and account are approved for this task?
- What information may go into it, and which connected sources may it access?
- Who checks the output and handles a mistake or unexpected disclosure?
If the answers are unclear, use a non-sensitive example while you resolve them. An approved tool still needs a defined task and accountable use. Our Everyday use resources offer a starting point for examining privacy, access and security choices.