Civil servants are steered to Microsoft Copilot as the Tax Agency and prison service tighten their rules
Sweden blocks ChatGPT and Claude on government work computers from 26 October 2026, citing security risks, TF finds.
Disclosure: Anthropic makes Claude, the AI assistant used to research and draft the article.
Microsoft Copilot is the sole general AI assistant approved. A small group of staff keep access through a platform the state built.
What’s Happening & Why It Matters
What the Ban Covers
The Government Offices of Sweden cut staff access to browser-based ChatGPT and Claude on work machines. Mikael Wettercrantz, director of digitalisation, told staff using AI services without a formal contract could harm information security. The Swedish Tax Agency and the National Board of Institutional Care restrict the tools too, according to Sveriges Radio. The Swedish Enforcement Authority is preparing a decision.
Some employees can still reach both tools through a government-built AI assistant platform. Reports do not say how many staff the exception covers or how many the ban affects.
Why Officials Acted
An audit at the Swedish Transport Administration earlier in 2026 found about 10 AI services in widespread and risky use. Officials fear staff paste sensitive data into consumer tools. Without a data agreement, providers could store inputs or use them for training. Reports say US law lets authorities compel American firms to hand over data.
Microsoft Copilot passed a security review and runs under an agreement between Microsoft and the Swedish government. ChatGPT and Claude received no comparable review. TF covered a similar US federal action against DeepSeek in US Bars DeepSeek From Federal Civilian Agencies.
An Open Question on Copilot
Microsoft has added Claude models to Copilot. In the EU, the Claude models default to off. Sweden has not said whether Claude runs inside its Copilot setup. OpenAI and Anthropic have not commented in the coverage TF reviewed.
The Ban’s Impact on Governments
Sweden’s move is a procurement decision, with no law behind the block. The pattern carries more weight than the single ban. Public bodies want enterprise contracts, audit rights and data residency before staff touch a chatbot. Vendors without such terms lose government seats.
My take: Blocking consumer chatbots at the door is a sound reflex, since unsanctioned tool use is the real exposure. The weak spot is the cure. Funnelling everyone into one vendor concentrates risk and bargaining power. Sweden should publish its review criteria so ChatGPT, Claude and others can qualify.
TF Summary: What’s Next
More Swedish agencies will issue rules, starting with the Enforcement Authority. OpenAI and Anthropic may offer enterprise terms to win government seats back. Other European governments will watch the Swedish review criteria.
MY FORECAST: Another EU government will restrict consumer chatbots on official devices by the end of 2026. OpenAI or Anthropic will announce a Swedish public-sector agreement by mid-2027. Sweden will publish approved-tool criteria within six months.
Related Stories
- US Bars DeepSeek From Federal Civilian Agencies
- Claude Bypasses Georgia Paywall as Elvis Estate Sues Anthropic
- OpenAI Cuts Frontier Model Training Off From the Live Internet
- Norway Expands Under-16 Social Media Ban to Games and Messaging Apps
- Nvidia Plans European Investment in UK, Stockholm and France