Sweden's government blocks ChatGPT and Claude, keeps Copilot
SummaryFrom Monday, October 26, staff at Sweden's Government Offices can't open the ChatGPT or Claude web apps on work computers. They get Microsoft Copilot instead. The difference is the contract and the settings, not necessarily the model.
· MODEL Editors · 2 min read
Some links may be affiliate links. Disclosure

Starting Monday, October 26, 2026, the people who support Sweden's prime minister and cabinet will type chatgpt.com or claude.ai on a work computer and get nowhere.
What happened. Sweden's Government Offices (Regeringskansliet, the Prime Minister's Office plus the ministries) is blocking the browser versions of ChatGPT and Claude on staff computers. Sveriges Radio's Ekot reported it first on October 23, and the Government Offices confirmed it in an email to SVT: "The browser interface for the AI services ChatGPT and Claude will no longer be reachable from the Government Offices' work computers from October 26."
- Why: Digitalisation director Mikael Wettercrantz wrote on the staff intranet that using AI services without a contract can have "consequences for our information security." He declined an interview with Ekot.
- What staff get instead: Microsoft Copilot, which the Government Offices approved and, after a review, made stricter. The press office won't say what was tightened, citing security.
- The exception: Some staff keep access to ChatGPT and Claude models through Intric, an AI assistant platform the Government Offices uses internally.
- Not the first: Ekot found similar blocks already in place at the Swedish Tax Agency and the National Board of Institutional Care (SiS), and the Enforcement Authority is preparing its own decision. An internal audit at the Transport Administration found widespread, risky use of about ten AI services.
The joke writes itself. Microsoft offers Anthropic's models inside Copilot. Notebookcheck notes they are off by default in the EU and have to be switched on by an admin, and the Government Offices hasn't said whether it did. So the ban is less "no Claude" than "no Claude without a contract."
Our take. This is the sensible version of an AI ban. The risk with consumer chatbots at work isn't which model answers. It's that a prompt typed into a personal account runs under the provider's default terms, with no agreement on retention, location or training. A contract with settings someone reviewed fixes that, whichever model is behind it. If your company still lets staff paste work into free chatbot accounts, Stockholm just showed you the cheaper fix: one approved door, with the settings checked.
What we did and didn't do: we read the Sveriges Radio, SVT, Dagens industri, Decrypt and Notebookcheck reports on October 25, 2026. We did not contact the Government Offices, and we don't know which models its Copilot setup has turned on. Not sponsored; no affiliate links.
Sources
- Regeringskansliet spärrar AI-tjänsterna Chat GPT och Claude (Oct 23, 2026)
Sveriges Radio Ekot
- AI-tjänster spärras på Regeringskansliet (Oct 23, 2026)
SVT Nyheter
- Regeringskansliet stoppar Claude och Chat GPT (Oct 23, 2026)
Dagens industri
- Sweden's Government Offices Block ChatGPT and Claude, Keep Microsoft Copilot
Decrypt
- Sveriges regering blockerar ChatGPT och Claude, behåller Copilot (Oct 25, 2026)
Notebookcheck
FREE · EVERY WEEK
The top AI tools this week, tested.
Every week MODEL's editors pick the AI tools worth your time, try them hands-on, and tell you what's good, what's not, and who it's for. Free, in your inbox.
Free. Unsubscribe anytime. Our picks are editorial and can't be bought.
The weekly three is editorial. A subscription does not buy a place in it.
Continue

Hot take
Aider's new CVE runs through shell=True lines Aider wrote itself
CVE-2026-115726 lets a cloned repo run shell commands on anyone who starts the Aider coding agent inside it. We read the git history: both shell=True calls the exploit goes through were authored by Aider. The real bug is that nobody asked who controls the string.
· MODEL Editors

Tech
Your old gaming GPU was doing AI math with the handbrake on
Open-source Linux kernel modules lift a half-rate limit on one kind of tensor math in GeForce RTX 20, 30 and 40 cards. An RTX 4090 doubles in that test, from 165.6 to 330.8 TFLOPS. Your games and your chatbot's typing speed stay about the same.
· MODEL Editors

Tech
AI model loyalty, enterprise edition: it's not you, it's your price per token
At Rippling, Anthropic models went from doing 57% of engineers' coding work in May to 18%, while OpenAI's share rose from 16% to 52%, The Information reports. The reason its CEO gave: Rippling wanted to cut costs, and OpenAI was cheaper.
· MODEL Editors