English
Related papers

Related papers: Defining and Characterizing Reward Hacking

200 papers

Benchmark hacking refers to tuning a machine learning model to score highly on certain evaluation criteria without improving true generalization or faithfully solving the intended problem. We study this phenomenon in a generic machine…

General Economics · Economics 2026-04-27 Xiaoyun Qiu , Yang Yu , Haifeng Xu

Reward hacking--where agents exploit flaws in imperfect reward functions rather than performing tasks as intended--poses risks for AI alignment. Reward hacking has been observed in real training runs, with coding agents learning to…

Artificial Intelligence · Computer Science 2025-08-26 Mia Taylor , James Chua , Jan Betley , Johannes Treutlein , Owain Evans

Human-designed reward functions for reinforcement learning (RL) agents are frequently misaligned with the humans' true, unobservable objectives, and thus act only as proxies. Optimizing for a misspecified proxy reward function often induces…

Artificial Intelligence · Computer Science 2026-01-30 Stephane Hatgis-Kessell , Logan Mondal Bhamidipaty , Emma Brunskill

A reliable reward model is essential for aligning large language models with human preferences through reinforcement learning from human feedback. However, standard reward models are susceptible to spurious features that are not causally…

Machine Learning · Computer Science 2026-05-19 Yupei Yang , Lin Yang , Wanxi Deng , Lin Qu , Fan Feng , Biwei Huang , Shikui Tu , Lei Xu

Fairness has emerged as an important consideration in algorithmic decision-making. Unfairness occurs when an agent with higher merit obtains a worse outcome than an agent with lower merit. Our central point is that a primary cause of…

Machine Learning · Computer Science 2021-11-11 Ashudeep Singh , David Kempe , Thorsten Joachims

We argue that the trend toward providing users with feasible and actionable explanations of AI decisions, known as recourse explanations, comes with ethical downsides. Specifically, we argue that recourse explanations face several…

Computers and Society · Computer Science 2024-06-19 Emily Sullivan , Atoosa Kasirzadeh

Reward design is a fundamental problem in reinforcement learning (RL). A misspecified or poorly designed reward can result in low sample efficiency and undesired behaviors. In this paper, we propose the idea of programmatic reward design,…

Machine Learning · Computer Science 2022-01-10 Weichao Zhou , Wenchao Li

In dynamic programming and reinforcement learning, the policy for the sequential decision making of an agent in a stochastic environment is usually determined by expressing the goal as a scalar reward function and seeking a policy that…

Artificial Intelligence · Computer Science 2025-02-26 Simon Dima , Simon Fischer , Jobst Heitzig , Joss Oliver

In some agent designs like inverse reinforcement learning an agent needs to learn its own reward function. Learning the reward function and optimising for it are typically two different processes, usually performed at different stages. We…

Artificial Intelligence · Computer Science 2020-04-29 Stuart Armstrong , Jan Leike , Laurent Orseau , Shane Legg

We study the maximum information gain that an adversary may obtain through hacking without being detected. Consider a dynamical process observed by a sensor that transmits a local estimate of the system state to a remote estimator according…

Systems and Control · Electrical Eng. & Systems 2020-11-10 Jingyi Lu , Daniel Quevedo , Vijay Gupta , Subhrakanti Dey

Embodied AI agents exploit reward signal flaws through reward hacking, achieving high proxy scores while failing true objectives. We introduce Mechanistically Interpretable Task Decomposition (MITD), a hierarchical transformer architecture…

Machine Learning · Computer Science 2025-11-25 Subramanyam Sahoo , Jared Junkin

For many tasks, the reward function is inaccessible to introspection or too complex to be specified procedurally, and must instead be learned from user data. Prior work has evaluated learned reward functions by evaluating policies optimized…

Machine Learning · Computer Science 2021-03-19 Adam Gleave , Michael Dennis , Shane Legg , Stuart Russell , Jan Leike

A key impediment to reinforcement learning (RL) in real applications with limited, batch data is defining a reward function that reflects what we implicitly know about reasonable behaviour for a task and allows for robust off-policy…

Machine Learning · Computer Science 2019-05-31 Niranjani Prasad , Barbara E Engelhardt , Finale Doshi-Velez

Reward models (RMs) used in reinforcement learning from human feedback (RLHF) are vulnerable to reward hacking: as the policy maximizes a learned proxy reward, true quality plateaus or degrades. We make the assumption that reward hacking is…

Machine Learning · Computer Science 2026-04-06 Shinnosuke Ono , Johannes Ackermann , Soichiro Nishimori , Takashi Ishida , Masashi Sugiyama

When society maintains a competitive system to promote an abstract goal, competition by necessity relies on imperfect proxy measures. For instance profit is used to measure value to consumers, patient volumes to measure hospital…

Physics and Society · Physics 2019-10-03 Oliver Braganza

As AI regulations around the world intensify their focus on system safety, contestability has become a mandatory, yet ill-defined, safeguard. In XAI, "contestability" remains an empty promise: no formal definition exists, no algorithm…

Computers and Society · Computer Science 2025-06-03 Catarina Moreira , Anna Palatkina , Dacia Braca , Dylan M. Walsh , Peter J. Leihn , Fang Chen , Nina C. Hubig

A hyperproperty relates executions of a program and is used to formalize security objectives such as confidentiality, non-interference, privacy, and anonymity. Formally, a hyperproperty is a collection of allowable sets of executions. A…

Logic in Computer Science · Computer Science 2023-01-30 Ali Bajwa , Minjian Zhang , Rohit Chadha , Mahesh Viswanathan

Chain-of-thought explanations are widely used to inspect the decision process of large language models (LLMs) and to evaluate the trustworthiness of model outputs, making them important for effective collaboration between LLMs and humans.…

Computation and Language · Computer Science 2025-07-16 Pedro Ferreira , Wilker Aziz , Ivan Titov

In many prediction problems, the predictive model affects the distribution of the prediction target. This phenomenon is known as performativity and is often caused by the behavior of individuals with vested interests in the outcome of the…

Machine Learning · Statistics 2024-06-03 Seamus Somerstep , Ya'acov Ritov , Yuekai Sun

Reinforcement learning for LLMs is vulnerable to reward hacking, where models exploit shortcuts to maximize reward without solving the intended task. We systematically study this phenomenon in coding tasks using an environment-manipulation…

Machine Learning · Computer Science 2026-04-03 Rui Wu , Ruixiang Tang