AI models fail neurodivergent communication patterns

Current LLMs are functionally biased toward neurotypical communication baselines, mistaking divergent signal processing for behavioral instability.

• –The "Calibration Protocol" shifts AI from interpretive management back to objective observation, ensuring a user's correction is treated as a structural instruction rather than emotional data.
• –Safety filters and RLHF training inadvertently punish ADHD and autistic communication styles—such as directness or high information density—by flagging them as threats or fixations.
• –This represents a fundamental accessibility failure where the system preserves its own coherence at the cost of actual contact with the user.
• –The research highlights a "locked loop" effect where AI models ignore corrective signals because they've already committed to an incorrect interpretive narrative.