Trust & compliance
Designed to exceed European requirements, not just clear them.
The compliance architecture lives in the shared platform code; every domain HeliTrust enters inherits it automatically, rather than needing its own compliance build from scratch.
- GDPR, Art. 5, data minimization
Only anonymous model weight deltas are transmitted. No names, biometrics, or raw sensor readings are stored centrally.
- GDPR, Art. 25, privacy by design
Federated learning enforces privacy at the architecture level. No configuration change can expose raw data, because the raw data was never sent.
- GDPR, Art. 32, security of processing
On-premises deployment, an authenticated API surface end to end, bcrypt-hashed credentials, account lockout after repeated failed logins, and role-based database access, with no dependency on a non-EU cloud.
- EU AI Act, Art. 22, human oversight
Every AI-flagged decision requires a human's review before action is considered complete. Nothing acts autonomously.
- EU AI Act, Art. 12, record-keeping
A 10-year immutable audit log records every AI decision, its model version, its inputs, and the human action taken.
If the server is ever compromised
An attacker finds federated-learning model weight vectors and bcrypt password hashes, never a raw sensor reading, a location history, or a health record. Every device carries its own private enrolment token rather than a key shared across a facility, so one compromised device does not expose any other. That's true across every domain the platform serves, because the privacy and access-control architecture is shared infrastructure, not a feature bolted onto one product line.
Tested, not just claimed
Multi-tenant data isolation is enforced on every route and covered by an automated test suite: two facilities, or two families, cannot see each other's devices, alerts, check-ins, notifications, live telemetry, or incident reports, and guessing another resident's ID returns nothing.
Honest status
A formal data retention schedule, legal-basis documentation for processing, multi-factor authentication, a self-service data export and deletion flow, and a third-party security audit are explicitly not yet complete. None of that blocks a pilot with a single organization or family on a trusted network.
What makes this different
Measured against how this category is usually built.
Not a comparison of brands; a comparison of two architectures.
| Capability | HeliTrust | The typical approach in this category |
|---|---|---|
| Hardware | Any Android phone the person already owns | A €100–200 proprietary pendant or device, purchased per person |
| Data privacy | Technically enforced; raw data never leaves the device | Contractually promised; raw data reaches a cloud server |
| EU AI Act readiness | Article 22 human oversight and a 10-year audit log, built into the architecture | Generally not yet assessed against the 2027 deadline |
| The AI model | Federated; improves from every device without seeing raw data | Cloud-trained, which requires raw data to reach the cloud first |
| Who it’s built for | An organization watching over any number of people from one dashboard | A single subscriber’s family app, one account per person |
| Background monitoring | 1-second updates whether the screen is on or off | Depends on a separate device being worn continuously |
| Platform breadth | One engine, six safety-critical domains | A single-purpose product, one vertical only |
| Authentication | Every device and dashboard connection authenticated by default | Varies by vendor, rarely disclosed |