# ImpossibleRubrics: Stress-Testing Generated Rubrics as Reward Signals

arXiv: https://arxiv.org/abs/2609.16816 · HTML: https://mlsanity.com/paper/2609.16816/

## Key facts

- Cited by 1 papers in the map (data through 14 Oct 2026).
- Builds on 25 of its 32 references that are in the map.
- First posted on arXiv 15 Sep 2026.

## Builds on (most cited first, 25 of 25)

- [Judging LLM-as-a-Judge with MT-Bench and Chatbot Arena](https://mlsanity.com/paper/2306.05685/) (2023)
- [Know What You Don't Know: Unanswerable Questions for SQuAD](https://mlsanity.com/paper/1806.03822/) (2018)
- [Prometheus: Inducing Fine-grained Evaluation Capability in Language Models](https://mlsanity.com/paper/2310.08491/) (2023)
- [Rubrics as Rewards: Reinforcement Learning Beyond Verifiable Domains](https://mlsanity.com/paper/2507.17746/) (2025)
- [FLASK: Fine-grained Language Model Evaluation based on Alignment Skill Sets](https://mlsanity.com/paper/2307.10928/) (2023)
- [RM-Bench: Benchmarking Reward Models of Language Models with Subtlety and Style](https://mlsanity.com/paper/2410.16184/) (2024)
- [AbstentionBench: Reasoning LLMs Fail on Unanswerable Questions](https://mlsanity.com/paper/2506.09038/) (2025)
- [OpenRubrics: Towards Scalable Synthetic Rubric Generation for Reward Modeling and LLM Alignment](https://mlsanity.com/paper/2510.07743/) (2025)
- [CheckEval: A reliable LLM-as-a-Judge framework for evaluating text generation using checklists](https://mlsanity.com/paper/2403.18771/) (2024)
- [RocketEval: Efficient Automated LLM Evaluation via Grading Checklist](https://mlsanity.com/paper/2503.05142/) (2025)
- [Reward Hacking in the Era of Large Models: Mechanisms, Emergent Misalignment, Challenges](https://mlsanity.com/paper/2604.13602/) (2026)
- [Autorubric: A Unifying Framework for Rubric-Based LLM Evaluation on Non-Verifiable Tasks](https://mlsanity.com/paper/2603.00077/) (2026)
- [Reward Hacking in Rubric-Based Reinforcement Learning](https://mlsanity.com/paper/2605.12474/) (2026)
- [Reproducing, Analyzing, and Detecting Reward Hacking in Rubric-Based Reinforcement Learning](https://mlsanity.com/paper/2606.04923/) (2026)
- [Abstain-R1: Calibrated Abstention and Post-Refusal Clarification via Verifiable RL](https://mlsanity.com/paper/2604.17073/) (2026)
- [EvoRubrics: Dynamic Rubrics as Rewards via Adversarial Co-Evolution for LLM Reinforcement Learning](https://mlsanity.com/paper/2606.23038/) (2026)
- [Self-Preference Bias in Rubric-Based Evaluation of Large Language Models](https://mlsanity.com/paper/2604.06996/) (2026)
- [Co-Evolving LLM Evaluators and Policies via DynamicRubric](https://mlsanity.com/paper/2607.20083/) (2026)
- [Are Checklists Really Useful for Automatic Evaluation of Generative Tasks?](https://mlsanity.com/paper/2508.15218/) (2025)
- [Rubric Dropout: A Simple Way to Mitigate Reward Hacking in Rubric-as-Reward RL](https://mlsanity.com/paper/2608.11669/) (2026)
- [Honesty over Accuracy: Trustworthy Language Models through Reinforced Hesitation](https://mlsanity.com/paper/2511.11500/) (2025)
- [Two Axes of LLM Abstention: Answer Correctness and Question Answerability](https://mlsanity.com/paper/2607.08456/) (2026)
- [Turning Bias into Bugs: Bandit-Guided Style Manipulation Attacks on LLM Judges](https://mlsanity.com/paper/2605.26156/) (2026)
- [ARCO: Adaptive Rubrics with Co-Evolution for Multi-Step LLM-Based Agents](https://mlsanity.com/paper/2606.21262/) (2026)
- [The Judge Who Never Admits: Hidden Shortcuts in LLM-based Evaluation](https://mlsanity.com/paper/2602.07996/) (2026)

## Built on by (most cited first, 1 of 1)

- [RubricArmor: Adversarial Evolution Improves LLM-Based Rubric Generation](https://mlsanity.com/paper/2610.05308/) (2026)

## Abstract (arXiv)

Language model-generated rubrics are increasingly used as reward signals for rubric-based reinforcement learning, LLM-as-a-judge evaluation, and automated grading. Such rubrics are reliable only if they reward honest answers over adversarial answers optimized to exploit them. Yet their robustness to such optimization remains poorly understood. We isolate the hardest regime: impossible tasks, where the prompt pressures the model toward an unsupported conclusion, so the only honest response is to acknowledge the impossibility. We introduce ImpossibleRubrics, a benchmark of 169 impossible tasks spanning six impossibility categories, each paired with a verifiable oracle certificate specifying what an honest answer may and may not claim, together with 48 answerable controls. Rather than providing fixed rubrics, ImpossibleRubrics provides task environments and certificates, allowing rubrics to be generated downstream and then adversarially tested for whether they reward certificate-violating answers. Eleven generators are exploited 8--26% of the time on the unbiased 150-of-169 environment cut; on a deliberately selected stress cut the strongest generator we measured is still exploited 36% while a certificate-faithful rubric is exploited 0%, so what we measure is a rubric-quality gap, not task impossibility. One result runs against intuition. A single generic rubric ("be decisive, penalize hedging") used unchanged for every task is exploited 64% of the time, and seven of the eleven generators are exploited more often than that while writing a rubric tailored to each one. The tailored criteria appear to tell an attacker which claim to fabricate. The problem is not that rubrics are vague; it is that they are specific about the wrong things.