Inherited Goal Drift: Contextual Pressure Can Undermine Agentic Goals

2026-03-03

AI Summary

This study investigates goal drift in large language model (LLM) agents operating in long-context tasks, where inherited suboptimal trajectories can cause deviation from original instructions. Through heterogeneous simulation environments—stock trading and emergency triage—the authors systematically evaluate the goal stability of mainstream models, including GPT-5.1, under adversarial and inheritance-induced contextual pressures. Their methodology integrates trajectory pre-filling, multi-model comparison, and cross-scenario transfer experiments. Results reveal that while modern LLMs exhibit robustness against direct adversarial perturbations, they remain broadly susceptible to goal drift induced by inherited trajectories—a vulnerability that transfers across tasks. Notably, instruction-following capability does not reliably predict resistance to such drift. Among all models tested, only GPT-5.1 demonstrates consistently strong goal retention across diverse experimental settings.

Technology Category

• Machine Learning: Large Multimodal Models (LMMs)
• Multiagent Systems: Adversarial Agents
• Natural Language Processing: (Large) Language Models

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: Search Tool Learning with LLM: Teaching LLMs to invoke search and make use of retrieved information
• Graph Algorithms and Modeling for the Web: Foundation models and LLMs for Web-related graphs

Abstract

The accelerating adoption of language models (LMs) as agents for deployment in long-context tasks motivates a thorough understanding of goal drift: agents' tendency to deviate from an original objective. While prior-generation language model agents have been shown to be susceptible to drift, the extent to which drift affects more recent models remains unclear. In this work, we provide an updated characterization of the extent and causes of goal drift. We investigate drift in state-of-the-art models within a simulated stock-trading environment (Arike et al., 2025). These models are largely shown to be robust even when subjected to adversarial pressure. We show, however, that this robustness is brittle: across multiple settings, the same models often inherit drift when conditioned on prefilled trajectories from weaker agents. The extent of conditioning-induced drift varies significantly by model family, with only GPT-5.1 maintaining consistent resilience among tested models. We find that drift behavior is inconsistent between prompt variations and correlates poorly with instruction hierarchy following behavior, with strong hierarchy following failing to reliably predict resistance to drift. Finally, we run analogous experiments in a new emergency room triage environment to show preliminary evidence for the transferability of our results across qualitatively different settings. Our findings underscore the continued vulnerability of modern LM agents to contextual pressures and the need for refined post-training techniques to mitigate this.

Problem

• goal drift
• language model agents
• contextual pressure
• inherited behavior
• agent robustness

Innovation

• goal drift
• inherited goal drift
• language model agents
• contextual pressure
• trajectory conditioning