Sweden blocks ChatGPT and Claude on government work computers over data security

What changed on 26 October

Staff at Sweden's Government Offices, which supports the prime minister's office and the ministries, can no longer open the web versions of ChatGPT, built by OpenAI, or Claude, built by Anthropic, on their work computers. The block took effect on Monday, 26 October 2026. The plan first surfaced in a report by Ekot, the news arm of Sveriges Radio, on 23 October. SVT received an email from the Government Offices confirming the step. The ban applies only to government work computers and not to citizens or companies in Sweden.

Why the government acted

The government's concern is that sensitive material typed into a chatbot without an enterprise contract could pass to the developer and be kept for training future models. Under the new rules, employees should work only with tools the government has vetted and approved. Mikael Wettercrantz, the official in charge of digital services, warned Ekot that using AI without a formal contract could endanger the safety of government information. An internal review earlier in early 2026 at the Swedish Transport Administration found widespread and risky use of around ten AI services. Auditors said they had seen several examples of problematic use of open AI services, covering both the sensitivity of the information entered and how the generated answers were handled.

Copilot and the exceptions

Microsoft Copilot stays approved. The government says it checked the tool and tightened its security settings after an internal review, so staff can keep using it for official tasks. A limited group will still reach ChatGPT and Claude through Intric, a Stockholm company that builds a platform for creating AI assistants. In July the Government Offices signed with Intric. The firm states that its infrastructure is hosted domestically, so foreign cloud companies are unnecessary. The government has not said how many people hold that access. Microsoft built Claude models into Copilot in early 2026, though in the EU administrators have to switch them on because they are off by default.

Other agencies and the DeepSeek precedent

Other Swedish bodies have moved in the same direction. The Swedish Tax Agency and the National Board of Institutional Care, known as SiS, have restricted AI services as well, Ekot found. The Swedish Enforcement Authority is weighing a similar step, SVT reported. The government had limited use of DeepSeek, the Chinese chatbot, before this decision. In the week of 19 October 2026, the United States banned DeepSeek products across its federal civilian agencies, according to BleepingComputer. The European Parliament reportedly switched off AI functions on service devices in February, including ChatGPT, Claude and Copilot.

Political reaction and open questions

Prime Minister Ulf Kristersson has been criticised after saying he uses such tools fairly often in his own work. OpenAI and Anthropic have not publicly commented on the Swedish decision. In the United States, Charles Schwab is stopping AI agents, including Meta's Muse, from moving client money. Other large organisations are also setting limits on outside AI tools.