A system that sets how many AI mistakes are okay before action starts.
Think of a phone plan with a data limit. A few AI mistakes are okay. Hit the limit, and the service gets a timeout.
It sets error limits by risk. It can trigger checks, slow the AI down, or pause it.
Hallucination
An AI Error Budget sets an acceptable limit for hallucinations and other errors.
AI Guardrails
AI Guardrails can get stricter or pause service when the budget runs out.
Model Calibration
Model Calibration gives better error estimates for setting the limit.