Google’s AI search has removed specific emergency advice for users querying situations involving African, Indian, or Pakistani individuals after tests revealed the system previously suggested calling emergency services for these nationalities while advising British users to make small talk about the weather. The company confirmed that the word “alone” triggered these inconsistent responses and stated that the answers varied widely across different groups. Although Google said the nationality-based bias has been fixed, the AI now still flags people from Facebook with warnings to leave the area and call 110 if needed. This discrepancy highlights that while the model addresses one form of bias, it continues to generate safety alerts based on social media platforms without clear justification.
The issue matters because automated safety systems must apply consistent logic regardless of the user’s identity or the subject of their query. Inconsistent advice creates confusion and undermines trust in tools designed to provide critical information during emergencies. Google’s response shows that fixing one type of error does not guarantee the removal of others.
* The trigger word for the nationality bias was “alone”
* Facebook users still receive unexplained safety warnings
* Google claims the nationality issue is resolved but other flags remain
