Protective AI, Not Surveillance AI
If AI is going to help people hold their own boundaries, it must earn that role without watching what they do.
Static rules cannot adapt to a person's life.
Vulnerability is patterned but personal. AI is proposed here for one narrow purpose: making voluntary boundaries easier to set well.
SafePredict — proposed
An on-device model that studies broad usage rhythms and may suggest protective periods. Example: “Your activity patterns suggest that 11:30 PM to 1:30 AM is a recurring high-risk period. Would you like SafeTime to include this period?”
Recommend, never decide
Every AI suggestion requires explicit user acceptance. Users may ignore, edit, or disable recommendations entirely without losing access to SafeTime.
What SafeTime will not do
- No content monitoring
- No screenshots
- No keystroke logging
- No browsing history collection
- No behavioral profiling for advertising
- No selling of personal data
- No third-party surveillance
SafeTime is designed to protect people — not to watch them.
On-device processing
Behavioral learning should happen locally, with the device as the boundary of the data.
Consent and control
Users can review, edit, export, or delete SafeTime data and turn learning off at any time.
Low-resource friendly
Models must be small enough to run on mid- and entry-tier devices, not just flagships.
SafeTime AI Principles
AI should strengthen users' ability to make choices, not make choices for them.
Personal behavioral data should remain on-device wherever technically possible.
Users should understand what the system observes and why it recommends what it recommends.
Protection is opted into, never imposed.
Essential communication must always remain available.
SafeTime supports intention; it does not shame, score, or penalize.
Claims should be tested by researchers outside the project.
Design must work across cultures, languages, and device tiers.
Autonomy safeguards
Any restriction must be user-initiated, reversible after a stated cool-off, and clearly explained before activation.
Clinical caution
SafeTime is not a clinical treatment and is not a substitute for professional support.
These principles describe the intended design of a research system. They will be revised as user research, ethical review, and technical validation progress.