TL;DR

MHCLG, the housing and local government department, has built a central platform called the AI Gateway that gives its digital teams one approved way to reach roughly 20 AI models. Teams no longer wire up their own connections, security setup or support model. Several projects already use it and more are being onboarded.

What it actually is

Underneath, it is Azure API Management exposing a standard interface. Teams working in either Azure or AWS can call through it, and so can applications and services rather than only people at keyboards — which is what lets AI sit inside a departmental service instead of beside it in a chat window.

Sam Harrison, a technical architect at the department, framed the problem as teams separately wiring up connections to different providers, producing duplicated effort, delays and standards that vary by team. The answer was a single reusable platform carrying shared controls and one onboarding route.

Deliberately, the platform is not tied to any one supplier. Models from several providers sit behind the same interface and further ones can be added.

Governance is the actual product

The Gateway only reaches vendors and models that have been approved, applies departmental security controls, and feeds audit and monitoring systems. Before it went live it passed the department’s own AI approval process plus technical review, data protection sign-off and penetration testing. Role-based access control and sign-in both reuse what the department already ran, and it connects to MHCLG’s AI Register.

It also shows the department what is being spent. Consumption, adoption patterns and costs are visible centrally — which is the thing most organisations discover they lack well after adoption has spread, rather than at the start.

Notably, no savings figures were published. Lower development time and cost are claimed as benefits without numbers attached.

Looking forward

This is the shape public-sector AI procurement is settling into: not one department buying one model, but a governed hatch through which approved models are consumed and counted. Councils watching Whitehall for a template have one here, and MHCLG is openly inviting other parts of government to ask about it. Expect the missing benefits data to be the question that follows.