On-device models are the safety inversion for children's AI
The child-safety backlash against cloud AI companions is now legal fact: California SB 243 (signed 2025-10-13, effective 2026-01-01) mandates disclosure and self-harm protocols; the FTC opened a 6(b) inquiry into seven chatbot makers (2025-09-11); Character.AI cut under-18s off open-ended chat (2025-11-24) after Garcia v. Character Technologies (2025-05-21) found a duty of care to minors. Bonsai-class on-device models invert the risk model — a private, parent-controlled companion whose data never leaves the device.
Why it matters (spine — health). Human health, measured honestly. Spec a child-safe, on-device (ternary) companion + a vendor safety-compliance checklist for ParKG; make 'nothing leaves the device' the product claim.
What would prove it wrong (falsifier). On-device child models can't meet the safety bar in testing, or regulation bans minor-facing AI companions outright regardless of where they run.
Live-status check + triangulation. The child-safety backlash against cloud AI companions is now legal fact: California SB 243 (signed 2025-10-13, effective 2026-01-01) mandates disclosure and self-harm protocols; the FTC opened a 6(b) inquiry into seven chatbot makers (2025-09-11); Character.AI cut under-18s off open-ended chat (2025-11-24) after Garcia v. Character Technologies (2025-05-21) found a duty of care to minors. Bonsai-class on-device models invert the risk model — a private, parent-controlled companion whose data never leaves the device. Source: Primary · CA SB 243 / Garcia v. Character Technologies (2025-10-13); see source link. Every post-cutoff figure was tool-fetched against a live source during the Kimi K3 sweep (2026-07-20); see sweeps/2026-07-20-VERIFIED_DIGEST.md for the full provenance and any corrections of record.
Spec a child-safe, on-device (ternary) companion + a vendor safety-compliance checklist for ParKG; make 'nothing leaves the device' the product claim.
domains: health · clinical-ai · measurement