Sweden's Government Offices, which covers the Prime Minister's Office and the country's ministries, is cutting off browser access to ChatGPT and Claude on staff computers starting Monday, October 26. Sveriges Radio first reported the move, and the agency confirmed it in an email to broadcaster SVT. Microsoft Copilot stays on.
The stated reason is information security, not anything about the models themselves. Mikael Wettercrantz, the Government Offices' director of digitalization, told staff on the internal intranet that using AI services without a formal agreement can put information security at risk. "We have seen several examples of problematic use of open AI services," he wrote, according to a report from Decrypt.
What keeps Copilot in bounds, according to the agency, isn't the underlying technology. It's the contract. The Government Offices says it reviewed Copilot and tightened its security settings before clearing it for continued use. OpenAI's and Anthropic's models aren't gone entirely, either. Staff can still reach them through Intric, an internal platform the government controls directly, SVT reports.
One detail is still murky. The Government Offices hasn't said whether Anthropic's models, which Microsoft also bundles into Copilot, are switched on for its staff. Across the EU, those models are off by default in Copilot and have to be turned on by an administrator, so the answer depends on a setting nobody's confirmed yet.
Not the first Swedish agency to pull the plug
Sweden's Government Offices is following, not leading. The Swedish Tax Agency and the National Board of Institutional Care, which runs youth homes and addiction treatment centers, already have similar blocks in place, SVT reports. The Enforcement Authority is working on its own decision. And an internal audit at the Transport Administration reportedly turned up around 10 different AI services in widespread, risky use, according to Decrypt.
The pattern isn't confined to Sweden. The European Parliament switched off AI features on work devices back in February, saying it couldn't guarantee the data security involved.
The underlying worry applies just as much outside government offices. With a free ChatGPT or Claude account, prompts get processed on the provider's own servers under its standard consumer terms. A business contract, by contrast, can lock down how long data is kept, where it's processed, and whether it ever feeds back into training the model. On regular consumer accounts, that training question comes down to a privacy toggle users have to find and flip themselves.
The stakes aren't hypothetical. A separate study cited by Notebookcheck found that chat data from nine different AI chatbots ended up reaching ad trackers, the kind of leakage a government ministry can't easily explain away.
For now, Swedish civil servants who want to use ChatGPT or Claude at work will have to do it through Intric instead of a browser tab, a narrower door than before, but not a closed one.
ADFiled by The AI Desk
Models, assistants and the companies and chips behind them, reported from what was released and what was claimed, with the difference kept clear.
More from this desk →
Be the first to comment
Join the argument. No password, just your email or a passkey.