Dashboard preview
- 18 approved parties are visible before the next round starts.
- 32 total rounds are already tracked in the shared history.
- Budget remaining is carried forward from the platform privacy-budget editor (0.84).
Train across sites that can never pool their records. Each round ships only the update, ends in a signed proof, and leaves auditors a bundle they can examine. When EU AI Act provenance enforcement begins in August 2026, you can show who trained on what, and under whose approval.
Active collaborators visible on the federation dashboard before the next round starts.
Round history stays visible so operators can compare drift, throughput, and review state.
Remaining privacy budget carried forward from the live privacy-budget editor.
Five steps. Every one carries who joined, who approved, and the proof of what aggregated.
Register a node and capture the provider, region, telemetry, and join token.
Approve the participant list before starting a new round.
Collect local model updates while source records stay inside each organization boundary.
Verify the aggregation proof and privacy-budget spend-down on one record.
Generate the audit export and hand the signed bundle to reviewers.
Every site contributes an update; every site keeps its records.
| Site | Health | Metric | Budget | Contribution | Failed check |
|---|---|---|---|---|---|
| North Clinic | healthy | 0.91 | 0.11 | 18% | none |
| West Hospital | healthy | 0.88 | 0.09 | 14% | none |
| Finance EU | review | 0.83 | 0.16 | 12% | latency |
| APAC Region | dropped | 0.71 | 0.22 | 0% | budget |
Approve participant roster
Collect encrypted local updates
Run secure aggregation
Verify privacy checks
Sign round packet
The figures here are illustrative seed data. In a pilot, every count reads from your own rounds.
Multiple clinical sites contribute local imaging updates while protected health information stays inside each hospital boundary.
Coordinators see round health, approvals, and proof on one release record.
Regional risk teams improve a shared model without centralizing transaction histories into a single warehouse.
Security and compliance reviewers can inspect proof before the next learning round starts.
Country or partner-specific nodes keep local policy boundaries intact while contributing to a shared objective.
Privacy budget spend-down stays understandable for non-cryptographers.
Read-only here. To run these rounds on your own sites and keep the weights they produce, start a pilot.