SAS Institute Specialist Urges Risk-Based AI Oversight

Companies need to match oversight to how they use AI, according to Josefin Rosén, a specialist in trustworthy and ethical AI at SAS Institute. She says too much review can slow work, while too little can allow serious errors through.

Rosén recommends tracking each AI use case, including its purpose, cost, owner and follow-up. Approving a model once does not determine how it should be used across an organization. A language model that summarizes internal meetings may need simpler checks than one that answers a tender request, which should be reviewed against the company’s requirements and commitments.

She also calls for ongoing monitoring, clear responsibility for responding to errors, and requirements for data storage and processing to be considered when systems are designed. A pre-launch check may not be enough if usage changes or a model’s results shift. Explainability can help employees judge when an answer is usable and when it needs closer review, but it does not replace human judgment.

Rosén cited SAS’s Data and AI Impact Report: The New Economics of Trust, based on IDC survey data from 2,699 decision-makers in 28 countries across banking, insurance, life science and the public sector. The report’s Trustworthiness Index combines five areas: data quality and governance, model governance and oversight, explainability and fairness, responsible AI policy, and audit and accountability. Organizations with the highest scores more often reported strong or high returns on AI investments than those with the lowest scores. The findings show an association; they do not establish that governance alone caused higher returns.