Transit identifies bias risk in an on-time prediction model that varies by route. Which management action best selects an AI risk treatment?

When transit identifies bias risk in an on-time prediction model that varies by route, management must select an explicit risk treatment—mitigate, accept, transfer, or avoid—based on municipal impact and risk appetite. That selection connects measurement of route disparities to a governed decision about which risk treatment to apply based on municipal impact and risk appetite.

Selecting a treatment such as transfer or exploit based on vendor contracts and data throughput uses vendor and throughput considerations, not a selection based on municipal impact and risk appetite, and exploit applies to positive opportunity rather than bias risk. Choosing a treatment such as mitigate or accept based on training loss curves and feature importance relies on technical training diagnostics, not on selecting a risk treatment based on municipal impact and risk appetite. Deciding a treatment such as avoid or accept based on model version history and deployment frequency relies on version and deployment metadata, not the governed selection of a treatment based on municipal impact and risk appetite.

Exam caveat: treatment choice should be revisited as ridership patterns and model updates change disparity profiles. Operational check: record the chosen treatment, owner, and next review date in the AI risk register entry for the on-time model, with a metric that shows whether mitigation (if selected) is moving route-level error gaps.