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.
Phase-Aware Mixture of Experts for Agentic Reinforcement Learning
School of Computer Science and Engineering, Southeast University, Nanjing, China(东南大学计算机科学与工程学院,南京,中国);College of Computing and Data Science, Nanyang Technological University, Singapore(南洋理工大学 computing and Data Science学院,新加坡)
A Tutorial on Cognitive Biases in Agentic AI-Driven 6G Autonomous Networks
i2CAT Foundation(i2CAT基金会);Hostelworld Group(Hostelworld集团);Technical University of Catalonia (UPC)(技术大学(加泰罗尼亚));Khalifa University of Science and Technology(卡里玛大学);ISI/ATH;University of Patras(帕特拉大学)
EndoCogniAgent: Closed-Loop Agentic Reasoning with Self-Consistency Validation for Endoscopic Diagnosis
School of Biological Science and Medical Engineering, Southeast University(东南大学生物科学与医学工程学院);Jiangsu Key Laboratory of Biomaterials and Devices, Southeast University(江苏省生物材料与器件重点实验室);State Key Laboratory of Digital Medical Engineering, Southeast University(国家数字医学工程重点实验室);The First Affiliated Hospital of Nanjing Medical University(南京医科大学第一附属医院);Department of Computer Science and Engineering, The Chinese University of Hong Kong(香港中文大学计算机科学与工程系);Jiangsu Provincial Joint International Research Laboratory of Medical Information Processing(江苏省联合国际医学信息处理联合实验室);Laboratory of Image Science and Technology, the School of Computer Science and Engineering, Southeast University(图像科学与技术实验室,东南大学计算机科学与工程学院)
The Law-Following AI Framework: Legal Foundations and Technical Constraints. Legal Analogues for AI Actorship and technical feasibility of Law Alignment
MILD: Mediator Agentic System with Bidirectional Perception and Multi-Layered Alignment for Human-Vehicle Collaboration
Department of Civil Engineering, McGill University(麦吉尔大学土木工程系);Systems Hub, The Hong Kong University of Science and Technology (Guangzhou)(香港科学与技术大学(广州)系统中心);Interdisciplinary Centre for Security, Reliability and Trust, University of Luxembourg(卢森堡大学安全、可靠与信任跨学科中心)
Paying Less Generalization Tax: A Cross-Domain Generalization Study of RL Training for LLM Agents
Meta Superintelligence Labs(Meta超智能实验室);FAIR at Meta(Meta的FAIR);Northwestern University(西北大学);The Ohio State University(俄亥俄州立大学);University of Pennsylvania(宾夕法尼亚大学)