Which practice is central to maintaining safe AI systems over time?

• A. Continuous monitoring and fairness/bias testing
• B. Infrequent audits
• C. Only initial training data quality
• D. Hardware upgrades only

Answer: (A)

Continuous monitoring and fairness/bias testing are essential because AI systems operate in real, changing environments. Data distributions shift, user behavior evolves, and models can drift away from the conditions they were trained for. By continuously tracking performance, outputs, and safety metrics, you can catch degradation early, trigger retraining, and adjust controls before harms arise. Fairness testing specifically checks that outputs remain non-discriminatory as the system encounters new data and contexts, helping prevent biased decisions over time. This ongoing oversight supports reliable, safe operation throughout the model’s life cycle, not just at deployment.

Relying on infrequent audits misses signs of drift or new biases that emerge after deployment. Focusing only on the quality of the initial training data neglects how data and usage can change, leading to unseen risks. Hardware upgrades affect speed or capacity but don’t address the quality, safety, or fairness of the decisions the system makes.