Which approach best addresses model risk and bias in the TELUS enterprise AI program?

• A. Rely solely on bias assessments during development.
• B. Implement bias assessment, fairness controls, security, audit trails, and ongoing monitoring with rollback capabilities.
• C. Ignore model governance after deployment.
• D. Limit monitoring to initial deployment.

Answer: (B)

Managing model risk and bias requires end-to-end, continuous governance across the AI lifecycle. Bias and fairness issues can arise long after deployment due to data drift, changing usage, or new contexts, so it isn’t enough to assess bias only during development. The strongest approach combines several safeguards: ongoing bias assessment and fairness controls to enforce equitable outcomes; robust security to protect data and model integrity; audit trails that record decisions, data lineage, and changes for accountability; and ongoing monitoring with rollback capabilities to intervene and revert to a safer or better-performing version if issues are detected. This creates a feedback loop that detects drift, enforces ethical and compliant behavior, and enables safe mitigation, which a one-time check, post-deployment neglect, or limited initial monitoring cannot provide.