Google removes AI health summaries after investigation finds dangerous flaws
Google has started disabling its AI-generated health summaries (“AI Overviews”) for certain medical queries after reporting found the feature returning misleading advice. In one example, a search for normal liver blood test ranges produced tables of enzyme values without the clinical context needed to interpret them, potentially giving false reassurance to people who should follow up with a clinician. The underlying issue isn’t just a single wrong number; it’s that medical interpretation depends on factors like demographics, symptoms, and the specific test panel.
The episode also highlights a broader reliability problem with LLM-style summarization in high-stakes domains. Even when a system is designed to “summarize trusted sources,” it can still flatten nuance, miss safety caveats, and present uncertain conclusions with an authoritative tone. Google says most AI Overviews are accurate and that clinicians review health-related responses, but this is another reminder that medical search results need strong guardrails, clear uncertainty signaling, and prominent “talk to a professional” guidance.