Sweden Bans ChatGPT and Claude Across Government Agencies
Sweden has banned public sector employees from using OpenAI's ChatGPT and Anthropic's Claude for government work, citing risks of data leaks and uncontrolled shadow IT. The decision follows internal reviews showing sensitive records being fed into unregulated commercial platforms without enterprise data safeguards.
Why Sweden pulled the plug on consumer chatbots
Officials worried that staff pasting public data into commercial chatbots could send sensitive state records to US servers for model training.
The clampdown began after an internal audit at the Swedish Transport Administration uncovered problematic uses of public chatbots in daily workflows. Without commercial agreements guaranteeing data isolation, state employees were essentially feeding unredacted inputs into models whose parent companies can retain and reuse prompts for training.
Mikael Wettercrantz, Chief Information Officer and Director of Digitization, warned that unrestrained shadow IT could trigger serious consequences for Sweden's information security. When workers use standard consumer accounts, compliance with European data protection regulations evaporates the moment records leave local borders to be processed in US data centers.
DeepSeek set the precedent, but Western tools followed
Sweden previously barred China-based DeepSeek over GDPR violations and fears of state-directed data access, a posture mirrored by US federal warnings about potential software vulnerabilities. Expanding the ban to OpenAI and Anthropic confirms that Stockholm views US-hosted proprietary models as a sovereignty hazard rather than singling out foreign adversaries.
The move highlights the regulatory tightrope facing European agencies trying to modernize administrative workflows while policing third-party exposure. As developers look for enterprise solutions, like how 1Password lets AI agents log in without seeing your passwords, government security chiefs remain wary of sending state data through any service without ironclad, audited boundaries.
Sweden's move is a sensible correction to two years of reckless chatbot adoption in public offices. Telling staff not to paste confidential data into consumer tools never works; an outright block backed by enterprise-negotiated exceptions is the only realistic way to prevent sensitive government data from quietly training the next generation of foundation models.
Microsoft Copilot survives under tighter leash
The blanket restriction does not wipe out generative software entirely. Microsoft Copilot remains approved for Swedish civil servants, though agency administrators have introduced tighter data controls following the internal probe.
Sweden will also maintain a fenced-off testing ground. A small number of vetted personnel will continue to test ChatGPT and Claude through a designated internal environment designed to prevent prompts from leaking back to consumer model databases.
What it means for public sector workers across Europe
The Swedish directive signals a wider European pushback against consumer AI accounts in public administration. Government workers in the EU should expect similar crackdowns on unvetted desktop tools, with agencies standardizing strictly on enterprise cloud agreements that guarantee zero training on user prompts and ensure data stays pinned within European boundaries.
Frequently asked questions
Why did Sweden ban civil servants from using ChatGPT and Claude?
Audits revealed employees pasting sensitive state information into unapproved commercial tools, risking data exposure, cross-border privacy violations under European rules, and unauthorized use of official data for model training.
Can any Swedish officials still access ChatGPT or Claude?
Yes, but only a restricted group of authorized employees working through a dedicated internal platform built specifically to isolate data from standard consumer servers.
Is Microsoft Copilot also banned in Sweden?
No. Copilot remains permitted for official work because enterprise contracts cover the service, though Swedish agencies have placed it under stricter administrative security controls.
Sources
- ChatGPT and Claude blocked by Swedish government over "problematic" security concerns | TechRadar. techradar.com, 2026. https://www.techradar.com/pro/security/chatgpt-and-claude-blocked-by-swedish-government-over-problematic-security-concerns