This week Sweden's Regeringskansliet — the Government Offices that serve the prime minister and the cabinet — blocked ChatGPT and Claude on staff work computers. The stated reason was not performance or cost. It was that sending government work to an outside AI service, without a contract governing where that data goes, 'can have consequences for our information security.' When a national government reaches that conclusion, it's worth asking what the actual fix is.
blocked on Government Offices work computers over information security
Swedish public bodies now restricting public cloud AI — Government Offices, Tax Agency, SiS
the direction of travel: an internal AI platform the government runs itself
Sweden's Regeringskansliet — the Government Offices supporting the prime minister and cabinet — has blocked ChatGPT and Claude on staff work computers. Per SVT and Sveriges Radio, digitalisation director Mikael Wettercrantz told employees the problem is using AI services 'without contracts,' which 'can have consequences for our information security.' Microsoft Copilot stays, with hardened security, and a small group keeps ChatGPT/Claude access only through an AI platform the Government Offices built and run themselves. The same restriction has landed at the Tax Agency and the State Institutional Board; DeepSeek was already limited.
Notice what the objection actually was. Not that the models are bad — the staff were plainly using Claude and ChatGPT. The objection is that government work was leaving the building for a service whose data handling they couldn't fully govern. That's not a Swedish quirk or an AI-safety debate; it's the structural fact of cloud AI: every prompt is a packet going somewhere you don't control. A contract narrows the risk. It doesn't remove the destination. The only thing that removes the destination is not having one.
Sweden's Government Offices blocked ChatGPT and Claude on work computers, citing information-security risk from using AI 'without contracts.' Same move at the Tax Agency and SiS.
When a national government won't send its work to a cloud AI, the issue isn't the model — it's that the data leaves the building at all. A contract narrows that; it doesn't close it.
Run the model on hardware you own, in your own building. If nothing leaves, there's no destination to govern, no contract to negotiate, and nothing to intercept.
Privacy and sovereignty are the same decision. A government that must guarantee where its information lives cannot outsource the computation to a building it doesn't own — and increasingly, it's deciding it won't. The same logic that applies to a cabinet applies to a law firm, a hospital, or a home.
On-premise.
AuraVision builds AI that runs entirely on hardware in your own building — Aura in the room, Farsight in the rack. There's no cloud destination to secure, no third-party contract to depend on, and nothing in transit to intercept. The fix a government is reaching for by building its own platform is the thing we ship by default.
Know someone who should see this? Send it to them.