A Google AI Summary Said a Surgeon Was Dead. He Was Seeing Patients That Week.
What happened
Google's AI Overviews service told anyone who searched for Claudio Zorzi in October 2025 that he was dead. He was not. Zorzi is a well-regarded specialist in orthopedics and traumatology at a hospital in Verona, Italy, and at the time the AI summary appeared he was alive and seeing patients. The people who encountered the claim first were not journalists or researchers. They were his patients and his family.
The error had a traceable cause. A family doctor with the same name had recently died in Trentino, a different region. Google's AI summary system conflated the two men, drew on the death notice, and attached it to the prominent surgeon without a verification step that would have caught the mismatch. The output was confident, prominently positioned at the top of search results, and factually wrong in a way that nothing in the system's generation process appeared to have checked before publication.
The fallout moved quickly from personal alarm to formal legal action. The hospital where Zorzi works issued a diffida, a formal legal warning, to Google, demanding removal of the inaccurate information and putting the company on notice for potential liability. Zorzi announced he would pursue legal proceedings seeking compensation for patrimonial and non-patrimonial damages, citing what his team described as unlawful conduct. The hospital's decision to formalize the complaint rather than wait for an organic correction reflects how seriously institutions now treat AI-generated falsehoods about living individuals.
The incident lands alongside a documented pattern of AI Overviews failures. The service has produced inaccurate results across a meaningful share of finance and health-related searches since its broad rollout, and earlier high-profile errors prompted public statements without preventing recurrence. What shifted in October 2025 is that the subject of the error was prominent enough, and the harm specific enough, that the institutional response moved directly to legal proceedings rather than a support ticket.
The deeper problem the incident surfaces is the absence of an auditable trail. The system that generated the death claim left no visible record of which sources it weighed, how it resolved the name collision between two men sharing a name in different regions, or who bore accountability for the output once it reached a real person's screen. A provable record of what a system retrieved, when it retrieved it, and how it arrived at a factual claim about a named living person would have made the error identifiable and correctable before it reached patients. Without that trail, disputed outputs get managed as incidents rather than examined as decisions.
Reported impact
- Affected parties
- Not publicly disclosed
- Harm type
- Not publicly disclosed
- Scale
- Not publicly disclosed
- Financial impact
- Not publicly disclosed
- Regulatory action
- Not publicly disclosed
Classification
Relevant governance controls
Governance control mapping is not available for this record.
- No controls mapped
Not publicly disclosed
Control mapping is analytical. It does not state that any control would have prevented the incident.
Sources and evidence
This record was researched and written by the Index. The event is also catalogued in the following database, which is listed for cross-reference.