Since late October, staff at Sweden’s Government Offices can no longer open ChatGPT or Claude on their work computers. The stated reason: information security, and services without an agreement. Staff are sent to approved tools, Microsoft Copilot among them, whose security was tightened after a review nobody will describe. And some staff keep ChatGPT and Claude, through an assistant platform the government built itself.
So Sweden chose a contract. The same kind of American models stay in the building; only the paperwork changed. Is an approved Copilot safer than a blocked ChatGPT, or just better documented? The part I would copy is that internal platform. It is a gateway: one door that decides which model sees which data, logs every call, and lets you swap the model behind it. Own that door and a contract is one option among several. Rent it, and the contract is your whole plan.