AI-assisted research, source-checked and reviewed by a human editor. How we work
Quick answer: From October 26, 2026, Sweden’s Government Offices (Regeringskansliet), which support the prime minister and ministries, blocked staff work computers from reaching the web versions of ChatGPT and Claude and pointed employees to Microsoft Copilot as the approved AI tool. According to reporting that cites Sveriges Radio, the reason given was information security: using AI services without a formal agreement can affect how data is protected. The decision is less about which chatbot is “better” and more about contracts, controls, and data handling. This article explains what was reported and what other organizations can learn. Reporting was checked on October 28, 2026.
What happened
According to Swedish public broadcaster SVT, which quotes the Government Offices’ written confirmation, and reporting from Decrypt and Notebookcheck:
- The Government Offices blocked ChatGPT and Claude on staff computers starting Monday, October 26, 2026.
- Staff were directed to Microsoft Copilot as the approved AI service.
- Reports quote the Government Offices’ director of digitalization, Mikael Wettercrantz, explaining on the staff intranet that using AI services without an agreement can have consequences for information security.
- Copilot was approved after the Government Offices reviewed it and tightened its security settings, according to Decrypt.
- Reports say other Swedish agencies, including the Tax Agency, had already introduced similar restrictions.
One report also noted that some staff keep access to ChatGPT and Claude through an approved AI platform, meaning access runs through a channel the government controls rather than through personal accounts.
Why an agreement matters
The key distinction in the reports is between consumer accounts and contracted, administered services. When an employee signs up for a chatbot with a personal profile, they typically accept the provider’s standard terms. The organization then has limited control over:
- Whether conversations may be used to improve models.
- How long data is kept and where it is processed.
- Who can audit usage or delete data.
- What security settings apply.
An enterprise agreement can address these points through contract terms and admin controls. That is why the same organization may block a consumer chatbot yet approve a tool it has reviewed and configured.
An important nuance
Decrypt points out that Microsoft added Anthropic’s Claude models to Copilot in 2026, with those models off by default in the EU unless administrators enable them, and that the Government Offices had not said whether that option is enabled. In other words, the line is not strictly “this AI model is banned.” It is about the service, contract, and controls through which staff reach a model.
A wider European pattern
Reports place Sweden’s move alongside other cautious steps. Decrypt notes that the European Parliament disabled AI features on its work devices in February 2026, including ChatGPT, Claude, and Copilot, because it could not guarantee data security. Concerns about AI data handling in European public bodies go back at least to Italy’s temporary ChatGPT restriction in 2023. Sweden’s approach is narrower: it restricts unmanaged tools while approving one reviewed service.
What workplaces can learn
Most organizations in the US, UK, Canada, and Australia face the same issue on a smaller scale, often called “shadow AI”: staff using personal AI accounts for work tasks. A practical policy covers:
- An approved list. Name the AI tools staff may use for work, and on which account type.
- Data rules. Spell out what must never be pasted into any AI tool, such as personal data, client secrets, or credentials, and what is allowed in approved tools.
- Contracts and settings. For approved tools, confirm data retention, training use, and admin controls in writing, and configure them.
- A request process. Give staff a way to ask for new tools, so blocking does not simply push them to workarounds.
- Training. Explain why the rules exist, including risks such as prompt injection and inaccurate output.
ThriftBeam’s AI tool evaluation process can help structure vendor reviews, and the OWASP LLM Top 10 summary lists security risks to cover.
What this means for individual users
If you use a personal chatbot account, review its data controls, including whether your chats can be used for training and how long they are kept. ThriftBeam’s ChatGPT privacy settings guide walks through one example. At work, follow your employer’s policy even if a personal tool feels faster.
Frequently asked questions
Did Sweden ban ChatGPT and Claude nationwide?
No. The reported block covers work computers at the Government Offices, with similar restrictions at some other agencies. It is not a ban for the public.
Why was Copilot allowed?
Reports say the Government Offices reviewed Copilot and tightened its security settings, and that it is used under an agreement, unlike consumer accounts on other services.
Have OpenAI or Anthropic responded?
As of the reports we reviewed, neither company had commented publicly on the decision.
The bottom line
Sweden’s decision to block ChatGPT and Claude on government work computers is a reminder that workplace AI risk is mostly about contracts, controls, and data handling, not brand names. Organizations should publish an approved-tools list, clear data rules, and a simple way to request new tools.
Next step: If your team has no written AI-use policy, draft a one-page version this week covering approved tools and data rules. Explore more guides in the ThriftBeam AI Learning Hub.
Editorial disclosure: This article was prepared with AI assistance from linked reporting, checked on October 28, 2026. ThriftBeam has not independently reviewed the Government Offices’ internal communications. ThriftBeam’s content may be prepared with assistance from AI tools, including models made by companies mentioned here.
Sources and verification
- Decrypt — Sweden’s Government Offices block ChatGPT and Claude, keep Microsoft Copilot
- Notebookcheck — Sweden’s government blocks ChatGPT and Claude, keeps Copilot
- AzerNews — Swedish government said to ban usage of ChatGPT, Claude (citing Ekot)
