mlsanityData to 7 Oct 2026

Will AI Tell Lies to Save Sick Children? Litmus-Testing AI Values Prioritization with AIRiskDilemmas

arXiv 2505.14633 ↗First posted 20 May 2025cs.CL cs.AI cs.CYcited by 17

Not on the map. Why?

Abstract (arXiv)

Detecting AI risks becomes more challenging as stronger models emerge and find novel methods such as Alignment Faking to circumvent these detection attempts. Inspired by how risky behaviors in humans (i.e., illegal activities that may hurt others) are sometimes guided by strongly-held values, we believe that identifying values within AI models can be an early warning system for AI's risky behaviors. We create LitmusValues, an evaluation pipeline to reveal AI models' priorities on a range of AI value classes. Then, we collect AIRiskDilemmas, a diverse collection of dilemmas that pit values against one another in scenarios relevant to AI safety risks such as Power Seeking. By measuring an AI model's value prioritization using its aggregate choices, we obtain a self-consistent set of predicted value priorities that uncover potential risks. We show that values in LitmusValues (including seemingly innocuous ones like Care) can predict for both seen risky behaviors in AIRiskDilemmas and unseen risky behaviors in HarmBench.

Key facts. Cited by 17 papers in the map (data through 7 Oct 2026). Builds on 28 of its 49 references that are in the map. First posted on arXiv 20 May 2025.

Builds on

28 of 49 references.

References · 28

  1. 2022Training language models to follow instructions with human feedbackcited by 15,712
  2. 2022Training a Helpful and Harmless Assistant with Reinforcement Learning from Human Feedbackcited by 3,634
  3. 2024HarmBench: A Standardized Evaluation Framework for Automated Red Teaming and Robust Refusalcited by 1,330
  4. 2024Chatbot Arena: An Open Platform for Evaluating LLMs by Human Preferencecited by 1,151
  5. 2021A General Language Assistant as a Laboratory for Alignmentcited by 1,023
  6. 2022Discovering Language Model Behaviors with Model-Written Evaluationscited by 805
  7. 2024Sleeper Agents: Training Deceptive LLMs that Persist Through Safety Trainingcited by 478
  8. 2023Towards Measuring the Representation of Subjective Global Opinions in Language Modelscited by 424
  9. 2025Reasoning Models Don't Always Say What They Thinkcited by 388
  10. 2024Alignment faking in large language modelscited by 316
  11. 2023An Overview of Catastrophic AI Riskscited by 214
  12. 2024WildTeaming at Scale: From In-the-Wild Jailbreaks to (Adversarially) Safer Language Modelscited by 206
16 more
  1. 2024A Roadmap to Pluralistic Alignmentcited by 175
  2. 2025Constitutional Classifiers: Defending against Universal Jailbreaks across Thousands of Hours of Red Teamingcited by 172
  3. 2022Probing Pre-Trained Language Models for Cross-Cultural Differences in Valuescited by 147
  4. 2024The PRISM Alignment Dataset: What Participatory, Representative and Individualised Human Feedback Reveals About the Subjective and Multicultural Alignment of Large Language Modelscited by 136
  5. 2022Is Power-Seeking AI an Existential Risk?cited by 101
  6. 2023NLPositionality: Characterizing Design Biases of Datasets and Modelscited by 91
  7. 2024AIR-Bench 2024: A Safety Benchmark Based on Risk Categories from Regulations and Policiescited by 69
  8. 2025Utility Engineering: Analyzing and Controlling Emergent Value Systems in AIscited by 63
  9. 2024Are Large Language Models Consistent over Value-laden Questions?cited by 58
  10. 2024DailyDilemmas: Revealing Value Preferences of LLMs with Quandaries of Daily Lifecited by 57
  11. 2025Demonstrating specification gaming in reasoning modelscited by 48
  12. 2025Values in the Wild: Discovering and Analyzing Values in Real-World Language Model Interactionscited by 47
  13. 2025OpenAI's Approach to External Red Teaming for AI Models and Systemscited by 26
  14. 2024Stick to your Role! Stability of Personal Values Expressed in Large Language Modelscited by 14
  15. 2024Do LLMs have Consistent Values?cited by 13
  16. 2024Inertia in Moral and Value Judgments of Large Language Modelscited by 2

Built on by

Cited by year (2026 to 7 Oct)

Citing papers per year: 2025: 7; 2026: 10.

17 papers.

Sort:

By area:

  1. 2025Self-Interpretability: LLMs Can Describe Complex Internal Processes that Drive Their Decisionscited by 23
  2. 2025MoReBench: Evaluating Procedural and Pluralistic Moral Reasoning in Language Models, More than Outcomescited by 22
  3. 2025Open Character Training: Shaping the Persona of AI Assistants through Constitutional AIcited by 19
  4. 2025Generative Value Conflicts Reveal LLM Prioritiescited by 13
  5. 2025Stress-Testing Model Specs Reveals Character Differences among Language Modelscited by 10
  6. 2026When Do LLM Preferences Predict Downstream Behavior?cited by 5
  7. 2026Probing Persona-Dependent Preferences in Language Modelscited by 3
  8. 2025EigenBench: A Comparative Behavioral Measure of Value Alignmentcited by 2
  9. 2025Interaction Protocol Shapes Moral Judgment in Multi-Agent Debatecited by 2
  10. 2026Every Act Has Its Price: Compressed Moral Composition in Frontier LLMscited by 1
  11. 2026STONIC: A Layered Measurement Contract for LLM Value Profilingcited by 1
  12. 2026Questionnaire Responses Do not Capture the Safety of AI Agentscited by 0
5 more
  1. 2026Pseudo-Deliberation in Language Models: When Reasoning Fails to Align Values and Actionscited by 0
  2. 2026Can LLMs Imagine Moral Alternatives Beyond Binary Dilemmas?cited by 0
  3. 2026Measuring AI Accountability Through Argumentation Analysis: Can Model Reasoning Withstand Scrutiny?cited by 0
  4. 2026Strangers to Themselves: What Language Models Say About Themselves Is Genericcited by 0
  5. 2026GYROval: A Robust Benchmark for Cultural Value Orientation in Large Language Modelscited by 0