Imagine you’re in your kitchen, rolling out dough for a pizza, and when you pull it from the oven the cheese slides off the crust. You wonder what to do to keep it glued in place, so you turn to the internet for help. You expect the search results to hand you exactly what you need, only to be confronted with fresh AI-generated summaries that suggest adding about 1/8 cup of non-toxic glue to the sauce to make it stick better.
That scenario recently played out for a user testing a new feature rolled out by a tech giant. Ten days ago, Google announced it would weave artificial intelligence into its popular search engine, the gateway to the web for billions of people. The result is small AI-crafted summaries that appear at the top of the screen and pull information from across the network. The promise was clear: they could take a lot of tedious work off a user’s plate.
Yet the new search experience hasn’t worked as smoothly as hoped. In the past week or so, numerous users in the United States reported that the latest AI feature amplifies misinformation, fabricates data or historical facts that never happened, and validates information that could be dangerous, even something as trivial as a pizza topping routine.
“Completely dangerous”
Experts point to a core issue: the AI system, which compiles summaries from vast internet sources, struggles to distinguish truth from falsehood. Its simplified answers can rely on low-quality or unreliable material. In the pizza and glue example, a comment from an eleven-year-old Reddit post by a user with the handle “fucksmith” could be treated as credible input by the system.
In recent hours, AI researchers, technology journalists, and everyday users have flagged major problems with the feature. They note claims as varied as the number of US presidents who were white and the notion that Barack Obama is a Muslim—a conspiracy idea popular on the far right that gained traction after amplification by public figures. Critics describe the situation as dangerous and worrying.
During a recent conference, the CEO stated that the AI summary feature, initially available only in the United States, will roll out more broadly over time. The plan is to reach more than a billion people who will rely on these AI-generated summaries by year’s end.
However, the surge of issues with AI Overviews could push the company to rethink the approach. A veteran tech journalist asked whether Google might pause the AI summaries for a week to reassess their safety and accuracy, reflecting a growing call for caution in deployment and risk management.