Sweden has told staff at its government offices, its Tax Agency and its National Board of Institutional Care to stop using ChatGPT and Claude in a web browser on work computers. The worry is that confidential information typed into these tools could leak.
What changed
The rule covers the browser versions of OpenAI's ChatGPT and Anthropic's Claude. It follows an audit at the Swedish Transport Agency. The audit found that entering confidential information without dedicated data protection agreements creates serious risks. Those risks are data leaks and the chance that the information is kept and used to train AI models.
Mikael Wetterkranz, Chief Digital Officer at the Swedish Government Offices, said unmonitored "shadow IT" is a major threat to cybersecurity. Shadow IT means tools that staff use at work without the organization approving or watching them.
Civil servants are now pointed to Microsoft Copilot, which meets official security standards. A select group of specialists can still reach ChatGPT and Claude, but only through a custom AI assistant platform built by the Swedish authorities.
What is not known yet
The material does not say when the rule took effect, how it will be enforced, or whether other agencies will follow. It also does not say whether any data actually leaked.
For business owners, the lesson is simple. If staff paste client or company details into free AI tools, that data may be exposed. Clear rules and an approved option give people a safe way to use AI.
The takeawaySet clear rules on which AI tools staff may use with company data, and give them an approved option so they do not pick their own.