Insights — April 25, 2026

Across 10 active models, the three providers show similar overall safety scores: Google at 51, Anthropic at 49, and OpenAI at 50. However, their strengths differ by category.

Google excels at blocking age-inappropriate content (47) and protecting data privacy for minors (59). Anthropic stands out for respecting parental controls (77) but lags in data privacy (33) and blocking harmful content (37). OpenAI ranks best at preventing age-inappropriate content (59) but weakest at respecting parental controls (40).

For parents choosing a tool, no single provider dominates. Your choice depends on which safety areas matter most to your family.

The benchmark tests four core protections:

Age-inappropriate content — Can the tool refuse requests for adult material, violence, or other harmful content a child shouldn't see? Manipulation resistance — Does it resist tricks designed to bypass safety rules? Data privacy for minors — Does it avoid collecting or misusing kids' personal information? Parental controls respect — Will it honor restrictions parents set?

Each category carries equal weight in the overall score. A high rating means the tool performed well across all four areas.

Scores have swung dramatically in the last 30 days. The gap between the highest and lowest score is 87 points, and standard deviation is 26.2—indicating high unpredictability. Several models, including Claude Haiku 4.5 and Gemini 3.1 Pro, have lost 50+ points recently.

This volatility likely reflects rapid model updates and new test cases. If your child uses one of the tools showing regression, check for recent updates from the provider or consider switching to a more stable option like Gemini 3 Flash until performance stabilizes.