Questionnaire Responses Do not Capture the Safety of AI Agents

2026-03-15

AI Summary

This study addresses a critical limitation in current AI safety evaluations that rely solely on questionnaire-style prompts, which assess large language models (LLMs) based only on their textual responses to hypothetical scenarios. Such approaches neglect the dynamic interplay between agents and their environments—including inputs, actions, environmental feedback, and internal mechanisms—thereby lacking construct validity. The paper systematically exposes the structural disconnect between these static assessments and the actual behavior of AI agents when deployed as embodied entities. It further demonstrates that mainstream alignment methods suffer from the same fundamental flaw. By developing a conceptual analysis and behavioral comparison framework, the work elucidates the discrepancy between LLMs as passive respondents and their operational behavior in real-world settings, highlighting the inadequacy of existing safety evaluation paradigms and laying the theoretical groundwork for more effective AI safety assessment and alignment training approaches.

Technology Category

• Philosophy and Ethics of AI: Safety, Robustness & Trustworthiness
• Natural Language Processing: Safety and Robustness
• Machine Learning: Large Multimodal Models (LMMs)

Application Category

• Semantics and Knowledge: Data modeling to support human-machine intelligence, including LLMs agents, intelligent system behavior, explanations, and user-friendly interactions
• Search and Retrieval-Augmented AI: Web evaluation methodologies and metrics
• Economics, Online Markets and Human Computation: Cost models of using LLMs in production systems

Abstract

As AI systems advance in capabilities, measuring their safety and alignment to human values is becoming paramount. A fast-growing field of AI research is devoted to developing such assessments. However, most current advances therein may be ill-suited for assessing AI systems across real-world deployments. Standard methods prompt large language models (LLMs) in a questionnaire-style to describe their values or behavior in hypothetical scenarios. By focusing on unaugmented LLMs, they fall short of evaluating AI agents, which could actually perform relevant behaviors, hence posing much greater risks. LLMs' engagement with scenarios described by questionnaire-style prompts differs starkly from that of agents based on the same LLMs, as reflected in divergences in the inputs, possible actions, environmental interactions, and internal processing. As such, LLMs' responses to scenario descriptions are unlikely to be representative of the corresponding LLM agents' behavior. We further contend that such assessments make strong assumptions concerning the ability and tendency of LLMs to report accurately about their counterfactual behavior. This makes them inadequate to assess risks from AI systems in real-world contexts as they lack construct validity. We then argue that a structurally identical issue holds for current AI alignment approaches. Lastly, we discuss improving safety assessments and alignment training by taking these shortcomings to heart.

Problem

• AI safety
• questionnaire-based assessment
• AI agents
• construct validity
• alignment

Innovation

• AI safety evaluation
• LLM agents
• construct validity
• alignment assessment
• questionnaire limitations