Nicholas Diakopoulos

AI agents increasingly enter practitioner workflows through delegated, multi-step tasks, such as data analysis, document review, coding, and summarization. Existing governance debates tend to emphasize provider-level technical governance, which steers general model behavior, and policy, which defines the boundaries of legitimate use. Both are necessary, but neither fully specifies how domain-specific norms should guide the intermediate choices agents make during task execution. This article develops runtime configuration as a meso-level, agent-facing governance mechanism for this operational gap. Runtime configuration refers to persistent, inspectable, and revisable instructions and supporting materials loaded at use time that specify decision authority, documentation and evidence-preservation duties, and conditions for human escalation. These artifacts bridge domain practice and agent execution. They translate situated normative commitments into agent-facing guidance while connecting that guidance to technical controls, work outputs, and human review. We illustrate the framework through a case study of investigative journalism, comparing three conditions: an unconfigured baseline and two configured conditions that guided agent runs on a public-records data task. Across the runs, the clearest differences associated with configuration concerned the conditions of delegation rather than substantive accuracy: The runs differed in escalation, provenance, workflow recoverability, and the visibility of consequential decisions. The aim of runtime configuration is not to replace model alignment, policy, expertise, or institutional accountability. Instead, it makes situated delegation more inspectable by translating normative domain commitments into operational guidance for agentic work.

Generative AI tools have proliferated the current media landscape. Newsrooms are adapting, referencing generative AI use for efficiency, productivity, creativity, and research purposes (Diakopoulos...

Regulators and policy makers around the world have endorsed risk-based approaches for governing the challenges and opportunities from Artificial Intelligence (AI). In theory, these assessments should protect the public from the harmful impacts of AI, thus safeguarding important societal values such as the realization of fundamental rights. However, as an anticipatory decision-making tool, the success and legitimacy of risk-based regulatory approaches depend critically on methods that help policymakers and societal stakeholders anticipate how AI may impact a diverse citizenry, the many different ways in which technological change may unfold, and the factors that lead to desirable and less desirable visions of the future. In this paper, we first point out critical flaws in existing assessment methods and regulatory approaches and consequently identify the need for a shift in current approaches to assessment approaches. We call on scholars, policy makers, and the technology industry to engage in anticipatory approaches that go beyond compliance and checklist exercises and acknowledge the responsibility they have for impacts that go beyond what is known. Second, we propose a concrete alternative empirical approach to address our concerns: Scenario-based Sociotechnical Envisioning (SSE). By using SSE, practitioners can make assessments more inclusive, more aware of the socio-technical context in which AI systems are deployed, and more prospective in terms of identifying impacts before they occur. We encourage scientists, policy makers and legislators to use SSE as one method in their assessment toolbox. Third, we anticipate possible objections to our approach and pro-actively discuss how these can be met.