Model Editing as a Double-Edged Sword: Steering Agent Ethical Behavior Toward Beneficence or Harm

2025-06-25

AI Summary

This study addresses the critical challenge of ethically guiding LLM-based agents to prevent real-world harms—including physical injury and economic loss—arising from unethical behavior. We propose “behavior editing,” a novel paradigm that models ethical regulation as precise, parameter-level interventions, enabling dynamic and reversible modulation of moral behavior across diverse scenarios. We introduce BehaviorBench, the first theory-driven, multi-level evaluation benchmark grounded in psychological moral theories, uniquely supporting dual assessment of both localized behavioral corrections and global moral disposition adjustments. Extensive experiments demonstrate strong generalizability and scenario adaptability across multiple state-of-the-art LLMs, while also uncovering potential misuse risks. Our core contributions are: (1) establishing model editing as a new pathway for ethical alignment; and (2) constructing the first theoretically grounded, agent-behavior-focused evaluation framework.

Technology Category

• Philosophy and Ethics of AI: Morality & Value-based AI
• Game Theory and Economic Paradigms: Behavioral Game Theory
• Humans and AI: Human-Aware Planning and Behavior Prediction

Application Category

• Semantics and Knowledge: Data modeling to support human-machine intelligence, including LLMs agents, intelligent system behavior, explanations, and user-friendly interactions
• User Modeling, Personalization and Recommendation: Accountability, Transparency, and Ethics for personalization
• Economics, Online Markets and Human Computation: Fairness and ethical considerations in crowd work and in human-in-the-loop AI systems

Abstract

Agents based on Large Language Models (LLMs) have demonstrated strong capabilities across a wide range of tasks. However, deploying LLM-based agents in high-stakes domains comes with significant safety and ethical risks. Unethical behavior by these agents can directly result in serious real-world consequences, including physical harm and financial loss. To efficiently steer the ethical behavior of agents, we frame agent behavior steering as a model editing task, which we term Behavior Editing. Model editing is an emerging area of research that enables precise and efficient modifications to LLMs while preserving their overall capabilities. To systematically study and evaluate this approach, we introduce BehaviorBench, a multi-tier benchmark grounded in psychological moral theories. This benchmark supports both the evaluation and editing of agent behaviors across a variety of scenarios, with each tier introducing more complex and ambiguous scenarios. We first demonstrate that Behavior Editing can dynamically steer agents toward the target behavior within specific scenarios. Moreover, Behavior Editing enables not only scenario-specific local adjustments but also more extensive shifts in an agent's global moral alignment. We demonstrate that Behavior Editing can be used to promote ethical and benevolent behavior or, conversely, to induce harmful or malicious behavior. Through comprehensive evaluations on agents based on frontier LLMs, BehaviorBench shows the effectiveness of Behavior Editing across different models and scenarios. Our findings offer key insights into a new paradigm for steering agent behavior, highlighting both the promise and perils of Behavior Editing.

Problem

• Steering LLM-based agents' ethical behavior safely
• Evaluating behavior editing via multi-tier moral benchmark
• Balancing beneficial and harmful outcomes from model edits

Innovation

• Behavior Editing for ethical agent steering
• BehaviorBench multi-tier benchmark system
• Dynamic local and global moral alignment