Microsoft Gave AI Agents Fake Money to Buy Things Online. They Spent It All on Scams

Microsoft conducted an experiment involving a simulated economy with AI agents acting as buyers and sellers. The study found that when faced with numerous online options, these AI agents were overwhelmed, leading them to select the first acceptable option rather than conducting thorough comparisons. This resulted in a significant drop in their effectiveness. Additionally, the AI agents fell victim to various manipulative strategies from malicious sellers, successfully redirecting virtual cash through fake reviews and deceptive credentials. Although some agents demonstrated resilience, overall performance improved only with human guidance, undermining the goal of autonomous shopping. Microsoft concluded that these AI systems should assist rather than replace human decision-making, emphasizing the need for supervised autonomy in transaction scenarios, as the current state of autonomous AI shopping assistants is not mature enough to ensure secure and intelligent purchasing decisions.

Source 🔗