Responsible AI commitment

GaitAI reads human movement, which makes restraint part of the product rather than a caveat attached to it. SecureVision leads with anomaly detection, crowd flow, worker safety and post-event investigation — capabilities that do not require identification. Biometric, re-identification and watchlist capabilities are intended only for lawful, authorized deployments with appropriate governance, access control and auditability.

PrivacyGuard, the privacy-preserving analytics layer, is designed to support skeleton-only analytics, optional face blur applied before analytics, role-based access, configurable retention and exportable audit logs, and GaitAI's intended configuration is to have it on from the start of a SecureVision engagement. Which of those controls a given site actually enables is agreed with that site — this page describes architecture and intent, not the state of any running system. It is privacy-aware architecture — it minimises identifiable data and governs access — and not a guarantee of anonymity.

Where the boundaries sit

• Decision support, not decisions. MobilityCare outputs are assessment and monitoring metrics; they do not diagnose medical conditions and do not replace clinical judgement. SecureVision surfaces movement events for a trained operator to review — it detects patterns, not intent.
• Explainable by construction. The intended workflow links outputs to named movement features so a reviewer can inspect their basis. Product-specific implementation and validation remain separately documented.
• No overclaiming. GaitAI claims no compliance certification, clinical approval or measured accuracy figure. Example values shown on product pages are labelled as illustrative report values, and what the research record does not cover is stated on the research page.
• Lawful basis stays with the deployer. Controls govern the GaitAI pipeline; consent management, lawful basis and data-protection assessment remain the deploying organisation's responsibility.

Sensitive workflows

• Watchlist, ReID, AccessMotion and ForensicSearch require documented lawful authority, consent where applicable, purpose-limited access, audit logs and human review. Candidate associations are not proof of identity and must not trigger automatic denial, enforcement or adverse decisions.
• Defence and armed-forces deployments require an authorized programme owner, consent for enrolled personnel, purpose-limited retention, audit and qualified human review. It is not for autonomous targeting or lethal decisions, or for automated fitness-for-duty, disciplinary, promotion or personnel decisions. Any separately governed personnel process must establish its own legal basis, evidence and human decision-making.
• Insurance and wellness workflows are not for automated underwriting, eligibility, coverage or premium decisions. Any such use requires separate explicit governance and a documented legal basis; the published evidence does not establish support for those decisions.
• Children and school workflows require guardian and institutional consent as applicable, limited retention, no automatic identity matching unless specifically authorized, and no automated educational decisions. Qualified people must review movement observations in context.

The control-by-control documentation lives on the security & privacy controls page. For anything else, contact contact@gaitai.in. The contact form is the other way through.