Child-Safe AI Must Be Proved, Not Promised

Children did not wait for a complete policy before encountering artificial intelligence. AI now appears in search results, learning platforms, homework assistance, image tools, games, phones and conversations that can feel private and personal. A system may be introduced as an educational assistant while also collecting sensitive information, shaping attention, supplying emotional advice or displacing the struggle through which learning develops.

The phrase ‘safe for children’ must therefore become an evidence claim rather than a marketing description. A youth mode, age gate, parental dashboard or list of prohibited outputs may be useful, but none establishes how the system behaves across long conversations, crisis disclosures, local languages, shared devices, adversarial prompts or ordinary classroom pressure.

Responsible institutions need a child-safety chain: defined benefit → age and context → evidence before exposure → bounded data and capability → youth-specific testing → accountable adult presence → crisis referral → incident response → remedy → independent review → suspension or exit. Protection begins before procurement and continues after deployment.

Reuters reported on 22 September 2026 that Tom Siegel, who founded Google's trust-and-safety team and later served as a vice president, had joined Common Sense Media's Youth AI Safety Institute as its first executive director. He warned that technology companies were repeating mistakes made during the social-media era and argued for independent youth-risk standards comparable in purpose to crash testing.…