Privacy and safety

Designed to support security review without identity surveillance.

GuardVision Edge turns observable behavior into reviewable evidence. It does not identify people, predict intent, or automate enforcement decisions.

PRIVACY_BOUNDARY

Identity out of scope

No biometrics

No facial identity matching, biometric embedding storage, watchlist matching, or cross-day person profile is part of the product contract.

On-device first

Edge runtime

The Android runtime is designed to process camera frames locally and export only the minimum event metadata needed for review.

Human review

Decision support

Alerts describe observable cues and recommended review steps. Operators remain responsible for context and response decisions.

Multi-cue safety

No single cue

Weak signals such as object proximity or camera avoidance cannot produce high-risk claims without supporting time, space, and motion evidence.

Operating principles

Privacy boundaries are product requirements, not afterthoughts.

The system is built around short-lived anonymous tracks, local inference, and explainable alerts for operator review.

Allowed and prohibited claims

The website, app, and operator reports must use the same language boundary.

Allowed

  • Detect observable behavior patterns such as dwell time, zone entry, line crossing, crowding, and rapid approach.
  • Use anonymous Track IDs within a session for motion and event analysis.
  • Generate explainable alerts that include time, zone, score band, and evidence summary.
  • Support operator review and deployment-specific risk tuning.

Not allowed

  • Identify dangerous people or predict criminal intent.
  • Recognize faces, infer protected attributes, or match biometric identities.
  • Automatically punish, detain, or report a person to law enforcement.
  • Treat a single weak signal as a definitive high-risk event.

Data lifecycle

Collect less, retain less, explain more.

MVP deployments should use the smallest dataset that still lets operators validate why an alert was produced.

01

Camera frames

Frames enter CameraX ImageAnalysis for immediate on-device processing.

Transient by default

02

AI observations

Detections, pose landmarks, object signals, and zone intersections are normalized into structured runtime observations.

Session-scoped unless tied to an event

03

Event metadata

Risk events store anonymous Track ID, timestamp, zone, reason list, score band, and operator review state.

Policy-bound local retention

04

Evidence media

Screenshots or clips are optional and should be captured only for configured high-priority events.

Disabled or time-limited by policy

Review alignment

Built for privacy review, mobile security review, and AI risk review.

These references guide the product posture. They are implementation alignment points, not legal certification claims.

OWASP MASVS

Use encrypted local storage, limited export surfaces, permission rationale, and clear mobile data handling boundaries.

NIST AI RMF

Evaluate false positives, false negatives, human oversight, explainability, and scenario coverage before field rollout.

Google Play readiness

Present camera permission rationale, privacy policy evidence, data safety copy, and a clear no-biometric-positioning statement.

Operator responsibilities

Safe deployment still depends on site policy and operator practice.

GuardVision Edge is a decision-support system. Each deployment should define local rules before alerts are used operationally.

  1. 01Post site notices where local law or policy requires camera disclosure.
  2. 02Define event retention windows and delete expired evidence.
  3. 03Limit access to event exports and review dashboards.
  4. 04Review false positives and tune thresholds before scaling deployment.
  5. 05Document escalation rules so alerts remain human-reviewed.

This page summarizes product design boundaries and engineering controls. It is not legal advice, a compliance certification, or a substitute for local legal review.