# AI Alignment at Your Discretion

arXiv: https://arxiv.org/abs/2502.10441 · HTML: https://mlsanity.com/paper/2502.10441/

## Key facts

- Cited by 20 papers in the map (data through 14 Oct 2026).
- Builds on 66 of its 127 references that are in the map.
- First posted on arXiv 10 Feb 2025.

## Builds on (most cited first, 30 of 66)

- [GPT-4 Technical Report](https://mlsanity.com/paper/2303.08774/) (2023)
- [Training language models to follow instructions with human feedback](https://mlsanity.com/paper/2203.02155/) (2022)
- [Judging LLM-as-a-Judge with MT-Bench and Chatbot Arena](https://mlsanity.com/paper/2306.05685/) (2023)
- [Direct Preference Optimization: Your Language Model is Secretly a Reward Model](https://mlsanity.com/paper/2305.18290/) (2023)
- [Deep reinforcement learning from human preferences](https://mlsanity.com/paper/1706.03741/) (2017)
- [Training a Helpful and Harmless Assistant with Reinforcement Learning from Human Feedback](https://mlsanity.com/paper/2204.05862/) (2022)
- [Constitutional AI: Harmlessness from AI Feedback](https://mlsanity.com/paper/2212.08073/) (2022)
- [Learning to summarize from human feedback](https://mlsanity.com/paper/2009.01325/) (2020)
- [Fine-Tuning Language Models from Human Preferences](https://mlsanity.com/paper/1909.08593/) (2019)
- [Chatbot Arena: An Open Platform for Evaluating LLMs by Human Preference](https://mlsanity.com/paper/2403.04132/) (2024)
- [Large Language Models are not Fair Evaluators](https://mlsanity.com/paper/2305.17926/) (2023)
- [Red Teaming Language Models with Language Models](https://mlsanity.com/paper/2202.03286/) (2022)
- [ChatGPT Outperforms Crowd-Workers for Text-Annotation Tasks](https://mlsanity.com/paper/2303.15056/) (2023)
- [Just Ask for Calibration: Strategies for Eliciting Calibrated Confidence Scores from Language Models Fine-Tuned with Human Feedback](https://mlsanity.com/paper/2305.14975/) (2023)
- [Open Problems and Fundamental Limitations of Reinforcement Learning from Human Feedback](https://mlsanity.com/paper/2307.15217/) (2023)
- [AlpacaFarm: A Simulation Framework for Methods that Learn from Human Feedback](https://mlsanity.com/paper/2305.14387/) (2023)
- [LLM Evaluators Recognize and Favor Their Own Generations](https://mlsanity.com/paper/2404.13076/) (2024)
- [Bias and Fairness in Large Language Models: A Survey](https://mlsanity.com/paper/2309.00770/) (2023)
- [From Crowdsourced Data to High-Quality Benchmarks: Arena-Hard and BenchBuilder Pipeline](https://mlsanity.com/paper/2406.11939/) (2024)
- [UltraFeedback: Boosting Language Models with Scaled AI Feedback](https://mlsanity.com/paper/2310.01377/) (2023)
- [Dealing with Disagreements: Looking Beyond the Majority Vote in Subjective Annotations](https://mlsanity.com/paper/2110.05719/) (2021)
- [TrustLLM: Trustworthiness in Large Language Models](https://mlsanity.com/paper/2401.05561/) (2024)
- [RLHF Workflow: From Reward Modeling to Online RLHF](https://mlsanity.com/paper/2405.07863/) (2024)
- [Annotators with Attitudes: How Annotator Beliefs And Identities Bias Toxic Language Detection](https://mlsanity.com/paper/2111.07997/) (2021)
- [AI Alignment: A Comprehensive Survey](https://mlsanity.com/paper/2310.19852/) (2023)
- [PKU-SafeRLHF: Towards Multi-Level Safety Alignment for LLMs with Human Preference](https://mlsanity.com/paper/2406.15513/) (2024)
- [Can Large Language Models Transform Computational Social Science?](https://mlsanity.com/paper/2305.03514/) (2023)
- [Evaluating the Moral Beliefs Encoded in LLMs](https://mlsanity.com/paper/2307.14324/) (2023)
- [A Roadmap to Pluralistic Alignment](https://mlsanity.com/paper/2402.05070/) (2024)
- [Aligning with Human Judgement: The Role of Pairwise Preference in Large Language Model Evaluators](https://mlsanity.com/paper/2403.16950/) (2024)

## Built on by (most cited first, 20 of 20)

- [Inference-Time Reward Hacking in Large Language Models](https://mlsanity.com/paper/2506.19248/) (2025)
- [Temporal Sparse Autoencoders: Leveraging the Sequential Nature of Language for Interpretability](https://mlsanity.com/paper/2511.05541/) (2025)
- [Generative Value Conflicts Reveal LLM Priorities](https://mlsanity.com/paper/2509.25369/) (2025)
- [Stress-Testing Model Specs Reveals Character Differences among Language Models](https://mlsanity.com/paper/2510.07686/) (2025)
- [CLASH: Evaluating Language Models on Judging High-Stakes Dilemmas from Multiple Perspectives](https://mlsanity.com/paper/2504.10823/) (2025)
- [Rigor in AI: Doing Rigorous AI Work Requires a Broader, Responsible AI-Informed Conception of Rigor](https://mlsanity.com/paper/2506.14652/) (2025)
- [Bridging Human and LLM Judgments: Understanding and Narrowing the Gap](https://mlsanity.com/paper/2508.12792/) (2025)
- [Decoding Alignment: A Critical Survey of LLM Development Initiatives through Value-setting and Data-centric Lens](https://mlsanity.com/paper/2508.16982/) (2025)
- [EVALUESTEER: Measuring Reward Model Steerability Towards Values and Preferences](https://mlsanity.com/paper/2510.06370/) (2025)
- [Reflect: Transparent Principle-Guided Reasoning for Constitutional Alignment at Scale](https://mlsanity.com/paper/2601.18730/) (2026)
- [The Consensus Trap: Dissecting Subjectivity and the "Ground Truth" Illusion in Data Annotation](https://mlsanity.com/paper/2602.11318/) (2026)
- [Principles Do Not Apply Themselves: A Hermeneutic Perspective on AI Alignment](https://mlsanity.com/paper/2604.10673/) (2026)
- [AI Alignment From Social Choice Perspectives](https://mlsanity.com/paper/2606.21550/) (2026)
- [OLAF: Towards Robust LLM-Based Annotation Framework in Empirical Software Engineering](https://mlsanity.com/paper/2512.15979/) (2025)
- [Democratic Preference Alignment via Sortition-Weighted RLHF](https://mlsanity.com/paper/2602.05113/) (2026)
- [Three Models of RLHF Annotation: Extension, Evidence, and Authority](https://mlsanity.com/paper/2604.25895/) (2026)
- [Understanding Annotator Safety Policy with Interpretability](https://mlsanity.com/paper/2605.05329/) (2026)
- [Theoretical Limits of Language Model Alignment](https://mlsanity.com/paper/2605.07105/) (2026)
- [An Instrument to Evaluate Governance Proposals: AI Policy Analysis at Scale](https://mlsanity.com/paper/2607.28094/) (2026)
- [Detecting Inconsistencies in Model Specifications with LLM-as-Verifier Reasoning](https://mlsanity.com/paper/2610.01847/) (2026)

## Abstract (arXiv)

In AI alignment, extensive latitude must be granted to annotators, either human or algorithmic, to judge which model outputs are `better' or `safer.' We refer to this latitude as alignment discretion. Such discretion remains largely unexamined, posing two risks: (i) annotators may use their power of discretion arbitrarily, and (ii) models may fail to mimic this discretion. To study this phenomenon, we draw on legal concepts of discretion that structure how decision-making authority is conferred and exercised, particularly in cases where principles conflict or their application is unclear or irrelevant. Extended to AI alignment, discretion is required when alignment principles and rules are (inevitably) conflicting or indecisive. We present a set of metrics to systematically analyze when and how discretion in AI alignment is exercised, such that both risks (i) and (ii) can be observed. Moreover, we distinguish between human and algorithmic discretion and analyze the discrepancy between them. By measuring both human and algorithmic discretion over safety alignment datasets, we reveal layers of discretion in the alignment process that were previously unaccounted for. Furthermore, we demonstrate how algorithms trained on these datasets develop their own forms of discretion in interpreting and applying these principles, which challenges the purpose of having any principles at all. Our paper presents the first step towards formalizing this core gap in current alignment processes, and we call on the community to further scrutinize and control alignment discretion.