Consumers More Likely to Lie to AI Than Human Agents

Researchers at Sun Yat-sen University found consumers were more likely to behave dishonestly with AI customer-service agents than with human workers. The study appears in the Journal of Business Research and is based on a series of lab and field experiments.

Across the experiments, participants lied, took advantage of pricing mistakes and falsely claimed discounts more often when they believed they were dealing with an AI agent rather than a person. In a field test, respondents also exaggerated results more often to gain extra rewards when the counterparty was an AI system.

The authors attribute the pattern to a lower fear of social judgment when people interact with machines. The paper uses the term ‘anticipatory face loss’ to describe the discomfort people expect when they foresee embarrassment or social disapproval, and reports that social judgment had a larger influence on behavior than feelings of guilt.

The research tested features that might change that dynamic. When AI agents displayed cues associated with social presence, such as simulated eye gaze, or when the agents signaled higher competence, dishonest behavior declined. The paper reports that eye contact cues and stronger ability signals made the agents seem more socially aware and increased consumers’ expectation of being judged.

Examples of misconduct observed in the experiments include submitting false discount claims, exaggerating performance outcomes to collect larger rewards and exploiting clear pricing errors. The authors measured these behaviors in multiple settings and compared the influence of anticipated embarrassment with internal moral feelings.

The study comes as companies expand use of automated agents for sales, support and transactions. A March 2025 Gartner report projected AI agents could resolve 80% of customer-service issues by 2029. A separate study at the University of Castilla found that robots with moderate human features, including facial expressions and eye movement, produced more positive responses, while highly realistic robots sometimes triggered discomfort associated with the uncanny valley.

The paper notes that adding social cues and clearer signals of competence to AI interfaces reduced dishonest behavior in the authors’ experiments. The authors present the results as a behavioral and design issue for firms deploying AI in customer-facing roles.