Microsoft’s new Bing AI, integrated into the search engine, is generating fabricated information in its responses, according to a detailed analysis published today by developer Dmitri Brereton. The blog post dissects the pre-recorded Microsoft demo and finds multiple instances where the AI confidently states falsehoods, from product specifications to financial data.

Brereton highlights that Bing AI produced a completely inaccurate pros-and-cons list for pet vacuums, attributing false features to a best-selling model. In another test, the AI generated a Mexico City itinerary with nightlife recommendations that included hallucinated venues and misattributed descriptions. Most damagingly, when asked to summarize Gap and Lululemon financial reports, the AI made up key metrics like operating margin and earnings per share.

Microsoft’s demo showcased the AI as a leap forward in search, but Brereton argues that the product is no better than Google’s Bard, which recently stumbled in its own ad. The analysis notes that Bing AI’s errors went largely unnoticed, while Google’s gaffe attracted widespread criticism.

On Hacker News, the post quickly garnered over 1000 points and 600 comments. Users shared their own experiences with Bing AI fabricating events (e.g., giving wrong Super Bowl winner) and questioned whether Microsoft knowingly released a flawed product to grab headlines. The debate now centers on whether AI search assistants can be trusted for factual queries.