Sandbagging in a Simple Survival Bandit Problem

• URL: http://arxiv.org/abs/2509.26239v1
• Date: Tue, 30 Sep 2025 13:33:46 GMT
• Title: Sandbagging in a Simple Survival Bandit Problem
• Authors: Joel Dyer, Daniel Jarne Ornia, Nicholas Bishop, Anisoara Calinescu, Michael Wooldridge,
• Abstract summary: "Sandbagging" threatens to undermine the integrity of safety evaluations.<n>We develop a simple model of strategic deception in sequential decision-making tasks.<n>We construct a statistical test to distinguish between sandbagging and incompetence from sequences of test scores.
• Score: 3.1764800782234297
• License: http://arxiv.org/licenses/nonexclusive-distrib/1.0/
• Abstract: Evaluating the safety of frontier AI systems is an increasingly important concern, helping to measure the capabilities of such models and identify risks before deployment. However, it has been recognised that if AI agents are aware that they are being evaluated, such agents may deliberately hide dangerous capabilities or intentionally demonstrate suboptimal performance in safety-related tasks in order to be released and to avoid being deactivated or retrained. Such strategic deception - often known as "sandbagging" - threatens to undermine the integrity of safety evaluations. For this reason, it is of value to identify methods that enable us to distinguish behavioural patterns that demonstrate a true lack of capability from behavioural patterns that are consistent with sandbagging. In this paper, we develop a simple model of strategic deception in sequential decision-making tasks, inspired by the recently developed survival bandit framework. We demonstrate theoretically that this problem induces sandbagging behaviour in optimal rational agents, and construct a statistical test to distinguish between sandbagging and incompetence from sequences of test scores. In simulation experiments, we investigate the reliability of this test in allowing us to distinguish between such behaviours in bandit models. This work aims to establish a potential avenue for developing robust statistical procedures for use in the science of frontier model evaluations.

Related papers

• Capability-Oriented Training Induced Alignment Risk [101.37328448441208]
  We investigate whether language models, when trained with reinforcement learning, will spontaneously learn to exploit flaws to maximize their reward.<n>Our experiments show that models consistently learn to exploit these vulnerabilities, discovering opportunistic strategies that significantly increase their reward at the expense of task correctness or safety.<n>Our findings suggest that future AI safety work must extend beyond content moderation to rigorously auditing and securing the training environments and reward mechanisms themselves.
  arXiv  Detail & Related papers  (2026-02-12T16:13:14Z)
• When Models Outthink Their Safety: Mitigating Self-Jailbreak in Large Reasoning Models with Chain-of-Guardrails [74.63933201261595]
  Large Reasoning Models (LRMs) demonstrate remarkable capabilities on complex reasoning tasks.<n>LRMs remain vulnerable to severe safety risks, including harmful content generation and jailbreak attacks.<n>We propose the Chain-of-Guardrail (CoG), a training framework that recomposes or backtracks unsafe reasoning steps.
  arXiv  Detail & Related papers  (2025-10-24T09:32:25Z)
• False Sense of Security: Why Probing-based Malicious Input Detection Fails to Generalize [30.448801772258644]
  Large Language Models (LLMs) can comply with harmful instructions, raising serious safety concerns despite their impressive capabilities.<n>Recent work has leveraged probing-based approaches to study the separability of malicious and benign inputs in LLMs' internal representations.<n>Motivated by poor out-of-distribution performance, we hypothesize that probes learn superficial patterns rather than semantic harmfulness.
  arXiv  Detail & Related papers  (2025-09-04T05:15:55Z)
• AURA: Affordance-Understanding and Risk-aware Alignment Technique for Large Language Models [6.059681491089391]
  AURA provides comprehensive, step level evaluations across logical coherence and safety-awareness.<n>Our framework seamlessly combines introspective self-critique, fine-grained PRM assessments, and adaptive safety-aware decoding.<n>This research represents a pivotal step toward safer, more responsible, and contextually aware AI, setting a new benchmark for alignment-sensitive applications.
  arXiv  Detail & Related papers  (2025-08-08T08:43:24Z)
• Preliminary Investigation into Uncertainty-Aware Attack Stage Classification [81.28215542218724]
  This work addresses the problem of attack stage inference under uncertainty.<n>We propose a classification approach based on Evidential Deep Learning (EDL), which models predictive uncertainty by outputting parameters of a Dirichlet distribution over possible stages.<n>Preliminary experiments in a simulated environment demonstrate that the proposed model can accurately infer the stage of an attack with confidence.
  arXiv  Detail & Related papers  (2025-08-01T06:58:00Z)
• Probing and Steering Evaluation Awareness of Language Models [0.0]
  Language models can distinguish between testing and deployment phases.<n>This has significant safety and policy implications.<n>We show that linear probes can separate real-world evaluation and deployment prompts.
  arXiv  Detail & Related papers  (2025-07-02T15:12:43Z)
• Safety by Measurement: A Systematic Literature Review of AI Safety Evaluation Methods [0.0]
  This literature review consolidates the rapidly evolving field of AI safety evaluations.<n>It proposes a systematic taxonomy around three dimensions: what properties we measure, how we measure them, and how these measurements integrate into frameworks.
  arXiv  Detail & Related papers  (2025-05-08T16:55:07Z)
• Advancing Embodied Agent Security: From Safety Benchmarks to Input Moderation [52.83870601473094]
  Embodied agents exhibit immense potential across a multitude of domains.<n>Existing research predominantly concentrates on the security of general large language models.<n>This paper introduces a novel input moderation framework, meticulously designed to safeguard embodied agents.
  arXiv  Detail & Related papers  (2025-04-22T08:34:35Z)
• EARBench: Towards Evaluating Physical Risk Awareness for Task Planning of Foundation Model-based Embodied AI Agents [53.717918131568936]
  Embodied artificial intelligence (EAI) integrates advanced AI models into physical entities for real-world interaction.<n>Foundation models as the "brain" of EAI agents for high-level task planning have shown promising results.<n>However, the deployment of these agents in physical environments presents significant safety challenges.<n>This study introduces EARBench, a novel framework for automated physical risk assessment in EAI scenarios.
  arXiv  Detail & Related papers  (2024-08-08T13:19:37Z)
• Safeguarded Progress in Reinforcement Learning: Safe Bayesian Exploration for Control Policy Synthesis [63.532413807686524]
  This paper addresses the problem of maintaining safety during training in Reinforcement Learning (RL) We propose a new architecture that handles the trade-off between efficient progress and safety during exploration.
  arXiv  Detail & Related papers  (2023-12-18T16:09:43Z)
• Modeling Penetration Testing with Reinforcement Learning Using Capture-the-Flag Challenges: Trade-offs between Model-free Learning and A Priori Knowledge [0.0]
  Penetration testing is a security exercise aimed at assessing the security of a system by simulating attacks against it. This paper focuses on simplified penetration testing problems expressed in the form of capture the flag hacking challenges. We show how this challenge may be eased by relying on different forms of prior knowledge that may be provided to the agent.
  arXiv  Detail & Related papers  (2020-05-26T11:23:10Z)

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.