AI-assisted internal adversarial review · 2026-08-25
What the family-product red and purple teams found
Direct answer. Several AI roles and cross-model systems reviewed Foxy, Everyday Hero and Primary Power adversarially. Their repository-verifiable findings triggered bounded claims, paused or removed prompts and executable release controls. The reviewers were AI systems acting on the owner's instructions. This is not independent professional review, certification, user research or commissioned expert approval.
Review JSON · Everyday Hero safety-control JSON · 84-item scenario register · Commissioning brief
Interim controls now in force
EHQ-SAFE-001
Safety and digital packs are paused
All 33 safety- or digital-content scenarios, including five digitally themed items stored outside the digital area, stay out of child-facing direct and automatic queues until qualified item-level review; the two tier-3 safeguarding scenarios are not queued but remain visible in the public source pending review.
EHQ-SAFE-002
No moral or safety labels
Results use neutral practice language and do not label a child, award moral status or present a score as assessment.
EHQ-SAFE-003
No return-pressure streak
Daily streak loss and days-in-a-row rewards are not shown or awarded.
EHQ-SAFE-004
Lost-child disclosure is narrow
Prompts direct a child to a visible fixed information, customer-service or security counter and limit details to a verified official helper when needed.
EHQ-SAFE-005
Caregiver-led practice
The product describes caregiver-guided practice and prioritises a short family discussion or rehearsal.
EHQ-SAFE-006
Every scenario has a public status
The 84-item register publishes a stable ID, risk tier, queue status, provenance status and item-level source field. Empty source_ids means unverified, not approved.
What remains open
not-completed
No AI participant counts as an educator, safeguarding professional, speech-language practitioner, language reviewer, accessibility assessor, legal adviser or research participant.
everyday-hero-quest
External review not completed
primary educator; child-safeguarding professional; digital-safety specialist; accessibility and caregiver usability review
foxys-learning-patch
External review not completed
early-childhood educator; speech-language professional; native Singapore Mandarin or Simplified-Chinese and Singapore Malay review; accessibility review
primary-power-quest-sg
External review not completed
P1-P3 subject educators; accessibility and caregiver usability review
Scope and rollback boundary
The review covered /foxys-learning-patch/, /everyday-hero-quest/, /primary-power-quest-sg/, /privacy, /methods/. It did not run child or caregiver studies and makes no efficacy, certification or compliance finding.