Researchers find a way to catch AI chatbots pretending to be human in online studies

Scientists discovered that large language models can now pass traditional tests designed to catch bots (like simple puzzles humans solve easily), so they developed a new test based on how human memory actually fails — something AI systems don't replicate naturally. This matters because online research studies need to know whether participants are real people or machines, and the old detection methods no longer work.

For years, researchers running studies online could catch bots with basic IQ tests or pattern recognition tasks. But LLMs got good enough to pass those, creating a real problem: if you're studying how humans make decisions, you can't tell anymore whether your data comes from actual humans or AI systems mimicking them. This work suggests there's still a structural difference you can exploit — the specific ways human brains fail at memory tasks, which machines don't naturally replicate. The practical effect is that online behavioral research might stay viable, at least for now, but it means researchers need to get smarter about their screening methods or accept that their data will be contaminated with machine responses they can't detect.