NGOs and international organisations face a specific dilemma with generative AI. Their teams need it — for drafting, translation, research and reporting — but their beneficiary and donor data is exactly the kind of information that must never leave a controlled jurisdiction. Blanket bans do not hold; staff simply use personal accounts. The answer is an architecture that makes the safe path the easy path.
Three layers that work today
The pattern we deploy has three layers: policy, gateway, and hosting. None of them is exotic, and together they let an organisation say yes to AI without saying yes to data exfiltration.
- Policy: an acceptable-use policy aligned with ISO 27001 and the EU AI Act, defining approved tools, prohibited data classes and human-review requirements — short enough that people actually read it.
- Gateway: all AI traffic passes through a central proxy that authenticates users, logs prompts, and pseudonymises personal data on the way out — names, IBANs, case numbers are masked before any model sees them.
- Hosting: models run either in a Swiss data centre or on-premises for the most sensitive workloads, so prompts and outputs remain under Swiss law.
What the gateway changes
The gateway is the piece most organisations are missing. With redaction at the proxy, a programme officer can ask an AI assistant to summarise a case file and the model receives placeholders instead of identities. The answer comes back, the placeholders are re-substituted locally, and the audit log shows exactly what left the building — which is, by design, nothing identifying.
Where to start
Start with the policy and an inventory of what staff already use — the results are always instructive. Then pilot the gateway with one or two teams on real workloads, measure, and expand. Most organisations can go from nothing to a governed, sovereign AI capability in a quarter, at a fraction of the cost of the data incident it prevents.
InterHyve designs and operates this stack — policy, redaction gateway and Swiss hosting — as part of its AI Governance practice. Ask us for the reference architecture.
Draft article for the site mockup — technical review by the InterHyve team pending before publication.
Talk to our engineersLet's look at your environment.
A one-hour assessment with a senior engineer. No sales script, no obligation.