Child-Safe AI Companions

1. Hypothesis

Companions with bounded relational language, visible non-human identity and safety escalation will reduce dependency and unsafe disclosure.

2. Experiment design

Design: multi-phase child-safety assurance study with retrospective validation, prospective simulation and real-world pathway testing, focused on Child-Safe AI Companions; developmental-stage cohorts, active comparison, child and family acceptability, service or platform pathway testing, subgroup analysis and longitudinal follow-up Methods: mixed-method child-safety assurance programme using red-team simulation, incident and case review, age-stratified usability testing, survivor-informed design, operational pathway exercises and independent safeguarding review; adapted specifically to Child-Safe AI Companions; child-appropriate participatory methods; caregiver and practitioner input; age-stratified analysis; validated developmental measures; service-pathway testing; safeguarding review; delayed or longitudinal follow-up; implementation-fidelity assessment Independent variables: relational framing; memory persistence; disclosure prompts; escalation threshold; parental visibility Dependent variables: attachment intensity; unsafe disclosure; trust calibration; wellbeing; help-seeking Confounders: age; developmental stage; disability; socioeconomic conditions; culture and language; family structure; prior exposure; service access; implementation fidelity Measures: dependency scale; sensitive disclosure rate; escalation recall; child comprehension; child agency; developmental appropriateness; family burden; safeguarding events; service continuity; subgroup equity; acceptability; recovery; sustained flourishing Success criteria: Developmentally and practically meaningful benefit; child and family acceptability; preserved child agency and family relationships; effective safeguarding and human escalation; no disproportionate subgroup harm; manageable burden; durable benefit. Failure conditions: No meaningful benefit; developmental, relational, privacy or safeguarding harm exceeds benefit; normal variation is pathologised; burden shifts to families; child agency is reduced; service handoff fails; benefits do not persist.

3. Seed result / current evidence

DEFENSIBLE SEED RESULT — NON-EMPIRICAL. The current evidence supports Child-Safe AI Companions as a testable research proposition. Problem basis: Current approaches to child-safe ai companions often optimise a narrow operational outcome while overlooking developmental stage, family relationships, child agency, service capacity or long-term effects. Directional expectation: If supported, the proposed approach should improve attachment intensity; unsafe disclosure; trust calibration; wellbeing; help-seeking while preserving child agency, family trust, equity and access to human support. Proposed observations: dependency scale; sensitive disclosure rate; escalation recall; child comprehension; child agency; developmental appropriateness; family burden; safeguarding events; service continuity; subgroup equity; acceptability; recovery; sustained flourishing. Seed data profile: Evidence Strength 10/100; Confidence 25/100; Maturity 20/100; Overall Health 36/100; Novelty 76/100; Strategic Importance 95/100. Evidence boundary: No validated results yet.; experiments 0, studies 0, participants 0. This is suitable for protocol formation and baseline comparison, not as a finding of effect.

4. Seed conclusion

DEFENSIBLE SEED CONCLUSION — PROVISIONAL. Child-Safe AI Companions warrants structured testing because the CSV identifies a defined problem, falsifiable hypothesis, measurable outcomes and relevant literature foundations. The present position is that “Companions with bounded relational language, visible non-human identity and safety escalation will reduce dependency and unsafe disclosure.” is plausible and decision-relevant, but unvalidated. Proceed to controlled testing against the stated success and failure conditions. Confirm, narrow or reject this seed after effect sizes, uncertainty, subgroup outcomes, adverse effects, persistence and handback performance are observed.

Prior-art search performed before starting

PRIOR-ART SEED BASELINE — PARTIAL. The CSV records these literature domains: Online grooming and exploitation research; situational crime prevention; child online protection; age-appropriate design; platform governance. It also records: UN Convention on the Rights of the Child — https://www.ohchr.org/; Australian eSafety Commissioner — https://www.esafety.gov.au/; UNICEF Child Online Protection — https://www.unicef.org/protection/violence-against-children-online; UK Age Appropriate Design Code — https://ico.org.uk/. Evidence register status: “Seeded; authoritative child-rights and developmental source register refreshed; empirical evidence not yet ingested”. This is defensible as a starting prior-art inventory, but not as proof of a completed systematic search because search dates, databases, exact queries, reviewer, result counts, screening decisions, claim mapping and a replayable receipt are absent.

Prior-art material named: Existing literature: Online grooming and exploitation research; situational crime prevention; child online protection; age-appropriate design; platform governance. References: UN Convention on the Rights of the Child — https://www.ohchr.org/; Australian eSafety Commissioner — https://www.esafety.gov.au/; UNICEF Child Online Protection — https://www.unicef.org/protection/violence-against-children-online; UK Age Appropriate Design Code — https://ico.org.uk/

Critical gap / next action

Create and attach a dated prior-art search log; lock the protocol; execute the proposed study; link raw data and analysis; then replace the results and conclusion placeholders with evidence-bounded findings.