Return-Weighted Risk for Navigating an Evolving AI Landscape: Reconnecting Risk with Value in AI Risk Management

Counts differ because each database indexes a different set of publications. We treat OpenAlex as the canonical count; Google Scholar is not shown (no API, and crawling it violates its ToS).

Organizations undertake AI activities to create value, yet often have to decide whether to continue or expand them while still learning what value they deliver and what risks they introduce. Changes in capabilities, configurations, users, costs, and alternatives can weaken the evidence supporting an earlier approval. We propose return-weighted risk (RWR) as an activity-level governance approach that connects an evolving value case with a risk case subject to appropriately independent assessment and challenge at each authorization decision. Three sequential tests organize that decision: whether the action satisfies applicable constraints, whether residual risk is permissible, and whether the current value case justifies taking that permissible risk. Intended value directs additional risk inquiry while a baseline assessment covers risks beyond the intended benefit. Feasible alternatives remain part of the comparison, and material change in purpose, evidence, or operating conditions can reopen authorization. A hypothetical small-business-lending example shows why a successful pilot may support further controlled learning without yet supporting expansion. The approach makes the basis for continuing action explicit; its effects on governance decisions and outcomes require empirical evaluation.