Person-centred infrastructure for governed health data and AI collaboration
Federated Computing: the connective architecture
Federated Computing connects. Governance determines how participation is authorised.
Open Health uses Federated Computing to coordinate computation, services and resources across independently governed organisations.
But connectivity does not confer authority.
Each participating institution retains authority over its resources and establishes the conditions under which they may be used. Open Health carries the relevant parts of those conditions into execution so that every requested interaction can be evaluated before it proceeds.
The technology supports three complementary layers.
Federated learning
Learn across distributed data
Models can learn across datasets held by different organisations without requiring those records to be pooled.
Federated coordination
Use complementary resources together
Data, models, computation and services held by different organisations can be brought into relation for an agreed purpose.
Federated governance
Keep institutional authority binding
Participating institutions establish the conditions under which their resources may be used. Admission evaluates each requested interaction against those conditions before it proceeds and records the decision as evidence.
From agreement to governed action
Governance begins with the collaboration itself.
Partners agree the purpose of the collaboration, who can participate, which resources can be used and which activities are permitted.
Where does authority come from?
Authority does not come from Open Health, a credential or an AI agent.
It comes from the legitimate institutional arrangements governing the collaboration, which may include consent, clinical duties, contracts, research governance, statutory authority or other appropriate bases.
Open Health represents and evaluates the relevant conditions of that authority when an interaction is requested.
Open Health then carries those agreed conditions into the technology.
1. Agree
Define the collaboration
Partners establish the purpose, roles, resources, permissions and responsibilities that govern the collaboration.
2. Express
Turn agreed conditions into executable rules
Relevant governance conditions are represented through a constitution, executable policy and capabilities linked to the approved collaboration.
3. Admit
Check before an activity proceeds
Before an organisation, person or AI agent performs an operation, governed admission checks whether that interaction is permitted for that participant, purpose and context.
4. Evidence
Record the decision
The system produces an explicit ALLOW or DENY decision and records the basis of that decision as signed, auditable evidence.
Governing AI agents
AI agents can retrieve information, invoke tools, perform analysis and produce new outputs. But technical capability does not give an agent institutional authority.
An AI agent acts under authority; it does not possess institutional authority of its own.
Open Health therefore separates three forms of control:
Admission1
May the agent perform this activity?
Admission determines whether a particular agent may perform a particular operation under an agreed authority, purpose and context.
Isolation
Can the agent bypass the decision?
Isolation prevents the agent from reaching protected data, credentials, signers or other privileged resources outside the governed route.
Containment
What can the agent do once admitted?
Containment restricts tools, files, network access, credentials and runtime behaviour during execution.
An admitted agent therefore cannot turn access to a tool, credential, model or intermediate result into permission to perform another operation. Each governed interaction remains bounded by the authority, purpose and scope under which it was admitted.
Together, these controls allow AI agents to contribute computational capability without inheriting the authority of the organisations whose resources they use.
Designed to scale
Pilots often operate within fixed arrangements.
Sustainable services need to accommodate new organisations, data, models, services and increasingly AI agents without rebuilding the entire collaboration each time.
Open Health treats scaling as a problem of governed evolution.
The aim is to allow collaborations to change while preserving organisational authority, accountability and control over distributed resources.