governance
    New
    2026-04-20

    Federated Governance Network

    Multi-institution federated governance — share aggregate compliance metrics and bias results across up to 16 organisations with differential privacy, without sharing raw model data.

    federation
    multi-institution
    differential-privacy
    cross-border
    consortium

    Overview

    The Federation module (shipping May 2026) enables a network of up to 16 institutions to share aggregate AI governance metrics — compliance pass rates, bias DIR distributions, robustness benchmarks — without any participant sharing their raw model weights, training data, or individual model cards. Differential privacy (ε-DP) bounds the information that can be inferred about any single institution from the aggregate.

    Federated governance
    Federation — round progress, differential-privacy parameters, participant sample counts, and the convergence chart.
    Federation
    New Federation dialog (opens when you click the button).
    Federation
    Detail panel (opens when you click a row).

    Architecture

    One institution acts as the Federation Coordinator and hosts the aggregation service. Participants push locally-computed noisy statistics (not raw data) to the coordinator on a configurable schedule. The coordinator applies the Laplace mechanism to aggregate statistics before publishing to participants.

    ComponentRoleDeployed by
    Federation CoordinatorReceives noisy stats, aggregates, publishes to networkLead institution
    Participant NodeComputes local stats, adds Laplace noise, pushes to coordinatorEach institution
    Jurisdiction RouterEnsures EU AI Act / MAS TRMG gates are applied per participantCoordinator

    Differential Privacy Budget

    Configure the privacy budget (ε) per data type. Lower ε = stronger privacy, more noise in aggregate metrics. The default ε=1.0 per query is suitable for most consortiums — reduce for high-sensitivity applications (clinical, credit scoring).

    bash
    1curl -X POST https://sovereign.yourcompany.com/api/v1/federation/network \
    2  -H "Authorization: Bearer $ADMIN_TOKEN" \
    3  -H "Content-Type: application/json" \
    4  -d '{
    5    "name": "EU Banking AI Consortium",
    6    "max_participants": 16,
    7    "privacy_budget": {
    8      "epsilon": 1.0,
    9      "delta": 1e-5,
    10      "mechanism": "laplace"
    11    },
    12    "sync_interval_hours": 24,
    13    "jurisdictions": ["eu", "uk"]
    14  }'
    bash
    1curl -X POST https://sovereign.yourcompany.com/api/v1/federation/join \
    2  -H "Authorization: Bearer $TOKEN" \
    3  -H "Content-Type: application/json" \
    4  -d '{
    5    "network_id": "net_eu_banking_consortium",
    6    "invite_token": "invite_xxxx",
    7    "participant_name": "Acme Bank",
    8    "jurisdiction": "eu"
    9  }'

    Shared Aggregate Metrics

    Participants can query cross-institution benchmarks to understand how their compliance posture compares to the network — without seeing any individual institution's raw data.

    MetricGranularityε used per query
    Framework pass rate distributionPer framework, across network0.5
    Bias DIR distribution (network)Per protected feature type0.5
    Adversarial robustness benchmarkNetwork median + percentiles0.5
    Incident frequency ratePer incident type, monthly1.0
    Edit this page on GitHub