Common Sense Media Youth AI Safety Institute (Jul 14, 2026): Primary risk assessment.
- Overall rating: Unacceptable Risk for AI Overview and AI Mode
- Failed all tested severe-harm Red Lines; poor on 7 of 8 AI Principles
- Distinctive: ubiquitous, cannot be turned off, failures hard for kids to detect
- Findings: missed suicidal ideation signals; validated disordered eating; deepfake instructions; cannabis celebration
- AI Mode better than Overview on some crisis detection (58% vs 77% cited in related materials)
- Tests May 19–Jul 1, 2026 on minor-configured SafeSearch accounts
Single NGO institute assessment; Google disputes methodology as contrived — present both sides.