Safe Evidence CompanionA Safe Harbor child-safety projectCONCEPT PROTOTYPE
← Safe Evidence Companion
Future features

What Safe Harbor may explore next.

A single place to understand possible future capabilities, why they matter, and what must be proven before they can move forward.

01
Current focusNOW

Make the core experience easier to understand

The immediate priority is a clearer, more accessible prototype—not collecting incident reports or launching automated safety decisions.

Plain-language guidance

Help a young person understand the next step without legal, technical, or forensic language.

Accessible interaction

Improve reading level, keyboard use, screen-reader behavior, contrast, and comprehension across ages and needs.

Safer feedback

Collect product critique without requesting names, allegations, screenshots, or real incident details.

02
Interactive learningCONCEPT UNDER REVIEW

AI Safety Coach

A practice space that could help young people recognize manipulation and rehearse safer choices using reviewed, entirely fictional situations.

Learn the warning signs

Explore suspicious messages, fake profiles, AI voice scams, manipulated images, phishing, and social engineering.

Practice a safer response

Pause, verify identity elsewhere, protect private information, leave, set a boundary, or ask a trusted person for help.

Adapt without addiction

Difficulty could respond to demonstrated understanding—without streaks, shame, fear, rankings, or engagement pressure.

Safety boundary

It would not inspect a child’s private messages or use real incidents as lessons. Testing remains research-gated.

03
Trust and transparencyRESEARCH-GATED

Trust Calibration Lab

A proposed lab for testing whether children and caregivers understand what the app knows, suspects, cannot determine, and is allowed to do.

Separate fact from inference

Make verified details, personal reports, uncertainty, and unknown information visibly different.

Explain authority

Show what the app may prepare or suggest and what only a person, school, platform, clinician, or authority can decide.

Test comprehension

Compare wording and interface designs to find blind trust, false reassurance, confusion, or unnecessary distrust.

Safety boundary

Synthetic research only. It is not a live assessment tool and is not approved for participant testing.

04
Early-warning researchRESEARCH-GATED

Child Safety Observatory

A proposed privacy-preserving system for identifying emerging online-risk patterns without tracking or profiling individual children.

Watch patterns, not people

Study minimum-count, non-identifying signals for new grooming tactics, scams, AI-generated abuse, and harmful platform features.

Keep uncertainty visible

Surface pattern-level alerts for qualified review—not accusations, identity claims, or automated enforcement.

Measure real usefulness

Track lead time, false alerts, missed threats, privacy failures, and the burden placed on human analysts.

Safety boundary

No real-child monitoring is active. Names, private messages, images, precise locations, and individual histories are out of scope.

05
Synthetic safety testingRESEARCH-GATED

Youth Safety Lab

A synthetic digital-twin laboratory that could test safety interventions without exposing real children or asking them to experience harmful scenarios.

Entirely fictional participants

Model age-appropriate online behaviors without recreating, identifying, or predicting any real child.

Test difficult scenarios safely

Explore bullying, grooming, sextortion, scam funnels, radicalization, and manipulative recommendations in containment.

Compare interventions

Measure which warnings, explanations, controls, and human-help pathways reduce simulated risk without creating new harm.

Safety boundary

No real-user trials or intervention claims are authorized by the synthetic lab.

06
Coordinated supportEXPLORATORY CONCEPT

Family Safety

A consent-led coordination concept that could help a young person, caregiver, school, counselor, platform, or authority share only what each genuinely needs.

Young person remains visible

Show who may receive information, what they would see, why it may help, and which choices remain with the young person.

Least disclosure

Prepare different, minimal summaries for different helpers instead of exposing one complete record to everyone.

Track responsibility

Record what was deliberately shared, who received it, what response followed, and what remains unresolved.

Safety boundary

It would not automatically contact family, schools, police, platforms, or other organizations.

07
Later, only if justifiedHOLD

Decide whether any tightly scoped pilot is responsible

A pilot is not the assumed destination. It would remain on hold until independent review and measurable evidence support proceeding.

Independent scrutiny

Require child-safety, trauma-informed, privacy, security, legal, and accessibility review.

Published limits

State what was tested, what remains unknown, and which claims the evidence does or does not support.

Exit and rollback

Define when work must pause or stop, how data would be deleted, and how participants would be informed.

KEEP EXPLORING

Features, next steps, and research