Google AI Overviews Provided Dangerous Advice Including Eating Rocks and Using Glue on Pizza

In May 2024, Google launched AI Overviews, an enhanced search feature that provided AI-generated summaries at the top of search results. Almost immediately after the public rollout, users began discovering and sharing screenshots of bizarre and dangerous advice generated by the system. Notable examples included telling users to eat rocks for minerals, suggesting adding glue to pizza to prevent cheese from sliding off, and recommending gasoline for spaghetti recipes. The erroneous advice stemmed from the AI system's inability to distinguish between legitimate sources and satirical content. The glue-on-pizza recommendation, for instance, was traced back to a joke comment from Reddit user 'fucksmith' made over a decade earlier. Similarly, other dangerous suggestions appeared to originate from parody websites, satirical forums, and joke responses that had been indexed by Google's search engine and subsequently treated as authoritative sources by the AI system. The incident gained widespread attention on social media platforms, with users sharing screenshots of the absurd recommendations under hashtags like #GoogleAIFail. Technology journalists and AI safety experts quickly highlighted the serious safety implications of an AI system providing potentially lethal advice with the apparent authority of Google's search platform. The examples demonstrated how AI systems could amplify and legitimize dangerous misinformation when they lack proper safeguards. Google responded by acknowledging the issues and stating that they were 'isolating cases' that didn't represent the system's overall performance. The company emphasized that AI Overviews generally provided high-quality information but admitted that the system had limitations, particularly with unusual queries and satirical content. Google implemented rapid fixes to filter out problematic sources and adjusted the types of queries that would trigger AI Overviews. Within days of the incident becoming public, Google significantly scaled back the rollout of AI Overviews, reducing the frequency with which they appeared in search results. The company also implemented additional quality controls and began excluding certain categories of content sources. Industry analysts noted that the incident highlighted the challenges of deploying generative AI at scale without adequate safety measures, particularly when integrating with critical information services that users rely on for health and safety guidance.