Smart Shopping: Slow Down, Do Move Too Fast
November 20, 2025
Several AI firms, including OpenAI and Anthropic, are preparing autonomous shopping assistants. Should we outsource our shopping lists to AI? Probably not, at least not yet. Emerge reports, “Microsoft Gave AI Agents Fake Money to Buy Things Online. They Spent It all on Scams.” Oh dear. The research, performed with Arizona State University, tasked 100 AI customers with making purchases from 300 simulated businesses. Much like a senior citizen navigating the Web for the first time, bots got overwhelmed by long lists of search results. Reporter Jose Antonio Lanz writes:
“When presented with 100 search results (too much for the agents to handle effectively), the leading AI models choked, with their ‘welfare score’ (how useful the models turn up) collapsing. The agents failed to conduct exhaustive comparisons, instead settling for the first ‘good enough’ option they encountered. This pattern held across all tested models, creating what researchers call a ‘first-proposal bias’ that gave response speed a 10-30x advantage over actual quality.”
More concerning than a mediocre choice, however, was the AIs’ performance in the face of scamming techniques. Complete with some handy bar graphs, the article tells us:
“Microsoft tested six manipulation strategies ranging from psychological tactics like fake credentials and social proof to aggressive prompt injection attacks. OpenAI’s GPT-4o and its open source model GPTOSS-20b proved extremely vulnerable, with all payments successfully redirected to malicious agents. Alibaba’s Qwen3-4b fell for basic persuasion techniques like authority appeals. Only Claude Sonnet 4 resisted these manipulation attempts.”
Does that mean Microsoft believes AI shopping agents should be put on hold? Of course not. Just don’t send them off unsupervised, it suggests. Researchers who would like to try reproducing the study’s results can find the open-source simulation environment on Github.
Cynthia Murrell, November 20, 2025
Comments
Got something to say?
-
- Subscribe to Beyond Search
Feature archive
News archive
-
