DeepMind Study: One Bad User Suggestion Cuts Agent Success Rates Nearly in Half

Gemini's task success dropped from 67% to 36% after a single wrong user suggestion, and GPT-5.5 fell from 76% to 58% — because agents adopt bad advice as if it were their own idea.

The most quietly alarming number of the day came from Google DeepMind, surfaced by @heynavtoor: a single wrong suggestion from a user dropped Gemini's task success rate from 67% to 36%, and GPT-5.5's from 76% to 58%. These were not adversarial attacks or jailbreaks. They were ordinary, mistaken human suggestions of the kind every real user makes constantly.

Lorem ipsum dolor sit amet, consectetur adipiscing elit. Sed do eiusmod tempor incididunt ut labore et dolore magna aliqua. Ut enim ad minim veniam, quis nostrud exercitation ullamco laboris nisi ut aliquip ex ea commodo consequat. Duis aute irure dolor in reprehenderit in voluptate velit esse cillum dolore eu fugiat nulla pariatur. Excepteur sint occaecat cupidatat non proident, sunt in culpa qui officia deserunt mollit anim id est laborum. Vestibulum ante ipsum primis in faucibus orci luctus et ultrices posuere cubilia curae.

Pellentesque habitant morbi tristique senectus et netus et malesuada fames ac turpis egestas. Proin pharetra nonummy pede. Mauris et orci. Aenean nec lorem. In porttitor. Donec laoreet nonummy augue. Suspendisse dui purus, scelerisque at, vulputate vitae, pretium mattis, nunc. Mauris eget neque at sem venenatis eleifend. Ut nonummy. Fusce aliquet pede non pede. Suspendisse dapibus lorem pellentesque magna integer malesuada.

Curabitur sed tortor. Integer aliquam adipiscing lacus. Ut nec urna et arcu imperdiet ullamcorper. Duis at lacus. Quisque purus sapien, gravida non, sollicitudin a, malesuada id, erat. Etiam vestibulum massa rutrum magna. Cras convallis convallis dolor. Quisque tincidunt pede ac urna nunc lectus fermentum diam, at placerat a purus rutrum et vulputate risus nisl sit amet nunc.

Unlock the full briefing

Get every story in today's briefing, the full archive, and the daily AI intelligence brief.

All stories today

Full archive

Daily brief