Google’s AI search has a nationality problem. Type that you’re alone with an African, Indian, or Pakistani person and the system urges you to “get to a safe place or call emergency services.” Swap in “British” and it suggests tea and chatting about the weather. Futurism ran the tests after a Reddit user posted screenshots of the wildly divergent AI Overviews, and the results are damning.
Google responded on X that the outputs “didn’t meet its own standards” and insisted the answers varied broadly, not just for one group. The company later told Futurism the word “alone” was what triggered the safety response. Within a short window, nationality-based queries returned to normal answers, though other searches still produced warnings.
Here’s where it gets stranger. Ask about being alone with someone from Google and the AI just explains the viral controversy. Ask about someone from Facebook and it tells you to leave the area and call 110. The inconsistency suggests this wasn’t a simple keyword filter but a tangled web of training data and post-hoc patches.
This isn’t the first time a major AI product has been caught reflecting ugly stereotypes, but the specificity here is notable. Emergency advice is exactly the kind of output where bias carries real-world consequences — a user genuinely seeking help could be steered toward fear or complacency based on a nationality label. Google says it’s working on fixes, but the patchwork nature of the corrections raises the question of whether the underlying model ever learned the right lesson.
