UK AI Security Institute finds weak real-world persuasion

blog.karthiktadepalli.com8 min read

The author argues that while laboratory studies show frontier AI models are exceptionally persuasive, real-world evidence from elections and influence operations reveals a shocking lack of effective AI persuasion. This disparity exists because evaluations test AI under ideal, information-dense conditions, whereas real-world persuasion must overcome hardened human skepticism and hostile information filters. Open article →