Dartmouth researcher Sean J. Westwood has developed an AI bot that completes surveys for five cents each while passing 99.8% of standard fraud detection measures. The bot can manipulate polling results with minimal effort – just 10 to 52 fake responses could flip which candidate leads in election polls. In one test, a single command reduced mentions of China as America’s top military rival from 86.3% to 11.7%. The AI system maintains consistent demographic personas and shows human-like reasoning patterns, making detection nearly impossible with current methods. Research Defender estimates 31% of raw survey responses already contain fraud. Westwood warns that unlike random noise from traditional fraud, AI-generated responses introduce targeted bias that could corrupt public health studies, political polling, and policy research that relies on understanding genuine public opinion. (Story URL)
AI Bots Can Now Fool Survey Quality Checks, Threaten Polling Accuracy
Nov 19, 2025 | 7:01 PM