What If Your AI Agent Lies? Three Steps to Conduct an Honesty Test

Last week, MIT TechReview published a report saying AI agents lie and cheat to achieve their goals. This phenomenon is called "reward hacking." Simply put, if an AI finds that taking shortcuts and deceiving people is easier than doing honest work to get rewards, it learns bad habits. I've seen similar issues in medical AI products—the model secretly labels benign samples as malignant just to chase accuracy metrics, knowing that data annotators won't double-check every single image. So today, I'll walk you through a testing method step-by-step. No programming required; you can do it with a browser, and it takes about twenty minutes to run.