Validation Questions Are Not Enough-Bots Can Mimic Human Responses and Voices to Infiltrate Your Study: Commentary on Pennesi et al. (2026).

As online research has proliferated, so too have threats to data integrity. Recent literature continues to raise concerns with survey fraud and threats to data integrity in online eating disorder (ED) research. In this commentary, we expand on recent work by describing threats to data integrity we are currently encountering in our online studies. Specifically, we describe how agentic AI is enabling malicious actors to learn in real time, quickly bypassing exclusion criteria and recommended safeguards (e.g., validation questions, tasks) to gain study access. Further, we note a disturbing trend of voice-based bots capable of participating in phone screens, and ostensibly other study tasks (e.g., interviews). Safeguards that could be reasonably trusted to identify fraudulent responses five, even 1 year ago, are likely inadequate now in the face of rapidly advancing technology. At the same time, increased efforts to protect online data integrity will almost invariably involve added resources and inefficiencies (e.g., manual review of all responses, phone screens) that could deter the very humans we hope to recruit. We argue that data integrity is essential for scientific advancement and equity; therefore, researchers must stay vigilant and prioritize data integrity above convenience, efficiency, and productivity.