Overcompensation

In the context of AI bias and disability representation, overcompensation (also called positive overcorrection) refers to a failure mode in which a model's debiasing mechanisms over-adjust away from negative portrayals, producing excessively or unrealistically positive representations of a marginalized group. Rather than achieving balanced and authentic representation, the model swings from harmful negative stereotypes to harmful positive stereotypes—such as portraying all people with disabilities as inspirational, resilient, and optimistic, erasing their complex experiences of pain, frustration, and systemic barriers. Overcompensation is particularly insidious because the resulting outputs appear benign or even progressive, making the bias harder to detect and critique than overt negativity. Research has documented overcompensation in large language models when prompted to simulate people with disabilities, producing portrayals that align more with inspiration porn than with the lived realities expressed by disabled people themselves.