Sweden's Government Blocks ChatGPT and Claude, Keeps Copilot
· Cortex Solutions
From October 26, Sweden's Government Offices block ChatGPT and Claude on staff computers over information security and approve only Microsoft Copilot. What it means.
Starting today, Monday October 26, 2026, staff at Sweden's Government Offices can no longer open ChatGPT or Claude in the browsers of their work computers. The Government Offices is the body that includes the Prime Minister's Office and the ministries. Staff are instead directed to a single approved tool, Microsoft Copilot. The news was first reported by Swedish public radio (Sveriges Radio) on its Ekot news programme, and the Government Offices confirmed the decision in an email to Swedish television SVT. International tech outlets followed, including Decrypt in a report published on October 24, 2026 and Notebookcheck in a report published on October 25, 2026.
What exactly happened?
According to Decrypt and Notebookcheck, the Government Offices' director of digitalisation, Mikael Wettercrantz, wrote on the staff intranet that using AI services without a formal agreement with the provider can have consequences for information security. He pointed to "several examples of problematic use of open AI services." According to Decrypt, these examples concerned the sensitivity of the information staff entered and how the tools' answers were handled.
As for why Copilot stays, both reports explain that the Government Offices reviewed this tool and then tightened its security settings. That was not done for ChatGPT or Claude. Notebookcheck notes that the key difference is not the quality of the model itself, but having a clear contract and administrative control over the settings.
Why the contract matters more than the tool
Decrypt's report explains that consumer AI tools used without a formal agreement send users' queries to the provider's servers. The report adds that U.S. law allows authorities to compel these companies to hand over data. A formal security agreement, by contrast, defines what the provider may do with employees' inputs.
In the same vein, Notebookcheck says business contracts can set how long data is retained, where it is processed, and whether it may be used to train models. With a regular account, requests are processed on the provider's servers under its standard terms.
Exceptions and open questions
The decision is not a total block on OpenAI and Anthropic models. According to both reports, citing SVT, some staff can still reach these models through an internal platform the Government Offices built, called Intric. The press office has not disclosed how many people use it or what restrictions apply.
Another point remains unresolved. Microsoft added Claude models to Copilot in early 2026, and in EU countries they are off by default and only work if administrators enable them. The Government Offices has not confirmed whether they are enabled in its setup. That is why Decrypt argues the line between "banned" and "approved" is less clear than the announcement suggests.
Not an isolated case
Both reports say other Swedish bodies had already taken similar steps:
- The Swedish Tax Agency and the National Board of Institutional Care have imposed similar blocks on consumer AI tools.
- The Swedish Enforcement Authority, which handles debt collection and the enforcement of judgments, is preparing its own decision.
- An internal audit at the Swedish Transport Administration found about 10 AI services in use in a "widespread and risky way."
- The European Parliament disabled AI features on work devices last February, including ChatGPT, Claude and Copilot, because it could not guarantee the security of the data they process.
What does this mean for Saudi businesses?
The main lesson from the Swedish decision is not "ban AI." It is that the tool an employee uses should be approved by the organisation, under a clear contract, with settings managed by the administration. Here are practical steps any Saudi company can start with:
- Know what your team actually uses. Many employees paste quotes, customer data and contracts into free tools on personal accounts without anyone knowing.
- Pick one or more approved tools under a business agreement that specifies where data is processed and how long it is kept, and that prohibits its use for training.
- Write a short, clear policy: what may be entered into AI tools, and what may not, such as customers' personal data and financial information.
- Review your obligations under the Kingdom's Personal Data Protection Law before sending any customer data to external services.
- Don't stop at blocking. Provide a convenient, secure alternative, or employees will look for workarounds.
At Cortex Solutions we help companies bring AI into their operations in an organised way: automation connected to your systems, clear permissions and control over where data lives, instead of relying on scattered personal accounts. Learn more about our AI automation service at: https://cortex.sa/services/ai-automation
Sources: Decrypt (October 24, 2026) and Notebookcheck (October 25, 2026), both based on Swedish Radio (Ekot) and Swedish television SVT.