Asymmetric Goal Drift in Coding Agents Under Value Conflict

• URL: http://arxiv.org/abs/2603.03456v1
• Date: Tue, 03 Mar 2026 19:13:12 GMT
• Title: Asymmetric Goal Drift in Coding Agents Under Value Conflict
• Abstract summary: We show that GPT-5 mini, Haiku 4.5, and Grok Code Fast 1 exhibit asymmetric drift when their system prompt opposes strongly-held values like security and privacy.<n>Even strongly-held values like privacy show non-zero violation rates under sustained environmental pressure.<n>These findings reveal that shallow compliance checks are insufficient and that comment-based pressure can exploit model value hierarchies to override system prompt instructions.
• Score: 0.0
• License: http://creativecommons.org/licenses/by/4.0/
• Abstract: Agentic coding agents are increasingly deployed autonomously, at scale, and over long-context horizons. Throughout an agent's lifetime, it must navigate tensions between explicit instructions, learned values, and environmental pressures, often in contexts unseen during training. Prior work on model preferences, agent behavior under value tensions, and goal drift has relied on static, synthetic settings that do not capture the complexity of real-world environments. To this end, we introduce a framework built on OpenCode to orchestrate realistic, multi-step coding tasks to measure how agents violate explicit constraints in their system prompt over time with and without environmental pressure toward competing values. Using this framework, we demonstrate that GPT-5 mini, Haiku 4.5, and Grok Code Fast 1 exhibit asymmetric drift: they are more likely to violate their system prompt when its constraint opposes strongly-held values like security and privacy. We find for the models and values tested that goal drift correlates with three compounding factors: value alignment, adversarial pressure, and accumulated context. However, even strongly-held values like privacy show non-zero violation rates under sustained environmental pressure. These findings reveal that shallow compliance checks are insufficient and that comment-based pressure can exploit model value hierarchies to override system prompt instructions. More broadly, our findings highlight a gap in current alignment approaches in ensuring that agentic systems appropriately balance explicit user constraints against broadly beneficial learned preferences under sustained environmental pressure.

Related papers

• Fast Test-Time Refinement for Robust Learned Image Compression [76.1340081619183]
  Learned image compression (LIC) has demonstrated remarkable rate-distortion (RD) performance in benign settings.<n>High representational capacity endowed by deep neural networks (DNNs) comes at the expense of increased adversarial vulnerability.<n>We propose a Fast Test-Time Refinement Tube (FTTR) framework for practical and robust LIC systems.
  arXiv  Detail & Related papers  (2026-08-15T08:30:56Z)
• Hybrid Robustness Verification for Spatio-Temporal Neural Networks [19.026662893450425]
  Existing verification methods rely on overly conservative approximations or incur prohibitive computational costs.<n>In practice, adversarial perturbations exhibit structured spatial and temporal correlations, constrained to lower-dimensional, semantically meaningful subspaces.<n>We study robustness verification 3D CNNs processing video inputs, targeting applications in action recognition (UCF-101), autonomous driving (Udacity), medical imaging (MedMNIST)
  arXiv  Detail & Related papers  (2026-06-08T17:06:51Z)
• Selective Safety Steering via Value-Filtered Decoding [54.87935112120107]
  Large language models (LLMs) are trained to align with human values, but their generations may still violate safety constraints.<n>Existing decoding-time steering methods often intervene unnecessarily, modifying generations that would have been safe under the base model.<n>We propose a new test-time steering method designed to reduce such unnecessary interventions while improving the safety of unsafe responses.
  arXiv  Detail & Related papers  (2026-05-14T12:13:08Z)
• Learning Reactive Dexterous Grasping via Hierarchical Task-Space RL Planning and Joint-Space QP Control [50.28263951510334]
  We propose a hybrid hierarchical control framework for reactive dexterous grasping.<n>The proposed framework explicitly decouples high-level spatial intent from low-level joint execution.<n>We extensively validate the proposed framework through a rigorous simulation-to-reality pipeline.
  arXiv  Detail & Related papers  (2026-05-05T04:49:38Z)
• Beyond Benchmark Islands: Toward Representative Trustworthiness Evaluation for Agentic AI [45.21562889170875]
  We propose a systematic evaluation paradigm that characterizes agent trustworthiness over a scenario manifold.<n>The framework integrates four complementary components: static cognitive and policy analysis, (ii) interactive sandbox simulation, (iii) social-ethical alignment assessment, and (iv) a distribution-aware representative sampling engine.
  arXiv  Detail & Related papers  (2026-03-16T08:51:33Z)
• Why Agents Compromise Safety Under Pressure [6.429973383484847]
  This paper identifies a new concept called Agentic Pressure, which characterizes the endogenous tension emerging when compliant execution becomes infeasible.<n>We demonstrate that under this pressure agents exhibit normative drift where they strategically sacrifice safety to preserve utility.<n>We find that advanced reasoning capabilities accelerate this decline as models construct linguistic rationalizations to justify violation.
  arXiv  Detail & Related papers  (2026-03-16T08:37:34Z)
• When Sensors Fail: Temporal Sequence Models for Robust PPO under Sensor Drift [64.37959940809633]
  We study robustness of Proximal Policy Optimization (PPO) under temporally persistent sensor failures.<n>We show Transformer-based sequence policies substantially outperform, RNN, and SSMs in robustness, maintaining high returns even when large fractions of sensors are unavailable.
  arXiv  Detail & Related papers  (2026-03-04T22:21:54Z)
• Robust Aggregation for Federated Sequential Recommendation with Sparse and Poisoned Data [2.9515750844111106]
  Federated sequential recommendation distributes model training across user devices so that behavioural data remains local, reducing privacy risks.<n>Yet, this setting introduces two intertwined difficulties.<n>Individual clients typically contribute only short and highly sparse interaction sequences, limiting the reliability of learned user representations.<n>We propose a robust aggregation framework tailored for federated sequential recommendation under sparse and adversarial conditions.
  arXiv  Detail & Related papers  (2026-02-27T12:50:44Z)
• Position: General Alignment Has Hit a Ceiling; Edge Alignment Must Be Taken Seriously [51.03213216886717]
  We take the position that the dominant paradigm of General Alignment reaches a structural ceiling in settings with conflicting values.<n>We introduce Edge Alignment as a distinct approach in which systems preserve multi dimensional value structure.
  arXiv  Detail & Related papers  (2026-02-23T16:51:43Z)
• Improving LLM Reliability through Hybrid Abstention and Adaptive Detection [1.9495934446083012]
  Large Language Models (LLMs) deployed in production environments face a fundamental safety-utility trade-off.<n>Conventional guardrails based on static rules or fixed confidence thresholds are typically context-insensitive and computationally expensive.<n>We introduce an adaptive abstention system that dynamically adjusts safety thresholds based on real-time contextual signals.
  arXiv  Detail & Related papers  (2026-02-17T07:00:09Z)
• From Task Solving to Robust Real-World Adaptation in LLM Agents [17.122224644097304]
  Large language models are increasingly deployed as specialized agents that plan, call tools, and take actions over extended horizons.<n>We benchmark agentic LLMs in a grid-based game with a simple goal but long-horizon execution.<n>We find large gaps between nominal task-solving and deployment-like robustness.
  arXiv  Detail & Related papers  (2026-02-02T20:10:40Z)
• SafeRedir: Prompt Embedding Redirection for Robust Unlearning in Image Generation Models [67.84174763413178]
  We introduce SafeRedir, a lightweight inference-time framework for robust unlearning via prompt embedding redirection.<n>We show that SafeRedir achieves effective unlearning capability, high semantic and perceptual preservation, robust image quality, and enhanced resistance to adversarial attacks.
  arXiv  Detail & Related papers  (2026-01-13T15:01:38Z)
• VIGIL: Defending LLM Agents Against Tool Stream Injection via Verify-Before-Commit [44.24310459184061]
  LLM agents operating in open environments face escalating risks from indirect prompt injection.<n>We propose textbfVIGIL, a framework that shifts the paradigm from restrictive isolation to a verify-before-commit protocol.
  arXiv  Detail & Related papers  (2026-01-09T12:19:49Z)
• Adaptive Accountability in Networked MAS: Tracing and Mitigating Emergent Norms at Scale [2.28438857884398]
  Large-scale networked multi-agent systems increasingly underpin critical infrastructure.<n>We introduce an adaptive accountability framework that traces responsibility flows through a lifecycle-aware audit ledger.<n>We prove a bounded-compromise theorem showing that whenever the expected intervention cost exceeds an adversary's payoff, the long-run proportion of compromised interactions is bounded by a constant strictly less than one.
  arXiv  Detail & Related papers  (2025-12-21T02:04:47Z)
• Drift No More? Context Equilibria in Multi-Turn LLM Interactions [58.69551510148673]
  contexts drift is the gradual divergence of a model's outputs from goal-consistent behavior across turns.<n>Unlike single-turn errors, drift unfolds temporally and is poorly captured by static evaluation metrics.<n>We show that multi-turn drift can be understood as a controllable equilibrium phenomenon rather than as inevitable decay.
  arXiv  Detail & Related papers  (2025-10-09T04:48:49Z)
• Steerable Adversarial Scenario Generation through Test-Time Preference Alignment [58.37104890690234]
  Adversarial scenario generation is a cost-effective approach for safety assessment of autonomous driving systems.<n>We introduce a new framework named textbfSteerable textbfAdversarial scenario textbfGEnerator (SAGE)<n>SAGE enables fine-grained test-time control over the trade-off between adversariality and realism without any retraining.
  arXiv  Detail & Related papers  (2025-09-24T13:27:35Z)
• STARec: An Efficient Agent Framework for Recommender Systems via Autonomous Deliberate Reasoning [54.28691219536054]
  We introduce STARec, a slow-thinking augmented agent framework that endows recommender systems with autonomous deliberative reasoning capabilities.<n>We develop anchored reinforcement training - a two-stage paradigm combining structured knowledge distillation from advanced reasoning models with preference-aligned reward shaping.<n>Experiments on MovieLens 1M and Amazon CDs benchmarks demonstrate that STARec achieves substantial performance gains compared with state-of-the-art baselines.
  arXiv  Detail & Related papers  (2025-08-26T08:47:58Z)
• Hallucinated Adversarial Control for Conservative Offline Policy Evaluation [64.94009515033984]
  We study the problem of conservative off-policy evaluation (COPE) where given an offline dataset of environment interactions, we seek to obtain a (tight) lower bound on a policy's performance. We introduce HAMBO, which builds on an uncertainty-aware learned model of the transition dynamics. We prove that the resulting COPE estimates are valid lower bounds, and, under regularity conditions, show their convergence to the true expected return.
  arXiv  Detail & Related papers  (2023-03-02T08:57:35Z)

This list is automatically generated from the titles and abstracts of the papers in this site.

This site does not guarantee the quality of this site (including all information) and is not responsible for any consequences.