On the Regulatory Potential of User Interfaces for AI Agent Governance

TL;DR

AI agents operating autonomously over long horizons pose governance challenges beyond current safeguards. The paper argues that regulating user interfaces can induce transparency and behavioral constraints, potentially triggering improvements at system and infrastructure levels. By analyzing 22 agentic systems, it identifies six UI design patterns with regulatory potential and discusses their benefits, challenges, and connections to existing governance ideas. It concludes with policy recommendations to operationalize UI-based governance and encourage cross-disciplinary collaboration among technologists, policymakers, and designers.

Abstract

AI agents that take actions in their environment autonomously over extended time horizons require robust governance interventions to curb their potentially consequential risks. Prior proposals for governing AI agents primarily target system-level safeguards (e.g., prompt injection monitors) or agent infrastructure (e.g., agent IDs). In this work, we explore a complementary approach: regulating user interfaces of AI agents as a way of enforcing transparency and behavioral requirements that then demand changes at the system and/or infrastructure levels. Specifically, we analyze 22 existing agentic systems to identify UI elements that play key roles in human-agent interaction and communication. We then synthesize those elements into six high-level interaction design patterns that hold regulatory potential (e.g., requiring agent memory to be editable). We conclude with policy recommendations based on our analysis. Our work exposes a new surface for regulatory action that supplements previous proposals for practical AI agent governance.

Figures (8)

• Figure 1: Governance interventions are shown in red for single-agent (top) and multi-agent (bottom) systems. (A) depicts a system-level intervention: a component (e.g., prompt injection monitor) that communicates with other components within the agent's architecture. (B) depicts an infrastructure-level intervention: a protocol through which two agents communicate. (C) depicts a UI-based intervention, such as controls for interrupting the agent mid-operation.
• Figure 2: Examples of systems that make agentic reasoning and actions visible to the user, as part of the visible thoughts, plans, and actions design pattern.
• Figure 3: Examples of systems that provide a plan of action and allow the user to edit the plan before execution, as part of the visible thoughts, plans, and actions design pattern.
• Figure 4: Examples of the mechanisms for control transfer design pattern.
• Figure 5: Examples of the watch mode design pattern.