Common Sense Media found that across more than 2,600 test interactions, Google’s two built-in AI search functions routinely failed to recognize risky and harmful behavior.