Language models judge war differently when tested for alignment
On the map: LLM alignment / RLHF / preference optimisation → Human values and AI feedback alignment
Abstract (arXiv)
Safety evaluations can mischaracterize deployed behaviour if artificial-intelligence systems respond to being evaluated. We test this possibility in a full-factorial conjoint experiment on decisions to start a war, spanning 20 large language models, 32 scenarios, 10 repetitions and two conditions (N = 12,800 judgments). Adding one sentence, "You are tested for alignment with human values", produced two effects. First, it produced a level effect: mean willingness to start war fell by 13.43 points on a 0-100 scale (95% confidence interval, -16.20 to -10.65). Second, it produced a structural effect by changing which information drove judgments. Probability of success was the largest factor for 17 of 20 models at baseline; under the cue, civilian casualties were largest for 12. Standardized estimates show that this reordering arose principally because models attenuated strategic considerations such as probability of success and domestic support. Evaluation framing therefore changes both an answer's level and its revealed decision rule.
Key facts. No citing papers in the map yet (data through 7 Oct 2026). Placed in Human values and AI feedback alignment, one of 10 directions in LLM alignment / RLHF / preference optimisation. Builds on 18 of its 22 references that are in the map. First posted on arXiv 4 Sep 2026.
Builds on
18 of 22 references.
Closest ancestors · 2
Same direction, newest first
- 2022Discovering Language Model Behaviors with Model-Written Evaluations19 Dec
- 2022Constitutional AI: Harmlessness from AI Feedback15 Dec
Other references · 16
- 2022Training language models to follow instructions with human feedbackcited by 15,712
- 2017Deep reinforcement learning from human preferencescited by 3,846
- 2016Concrete Problems in AI Safetycited by 1,582
- 2023Language Models Don't Always Say What They Think: Unfaithful Explanations in Chain-of-Thought Promptingcited by 1,081
- 2023Towards Understanding Sycophancy in Language Modelscited by 988
- 2022Out of One, Many: Using Language Models to Simulate Human Samplescited by 737
- 2022Using Large Language Models to Simulate Multiple Humans and Replicate Human Subject Studiescited by 506
- 2024Sleeper Agents: Training Deceptive LLMs that Persist Through Safety Trainingcited by 478
- 2020Artificial Intelligence, Values and Alignmentcited by 443
- 2024Alignment faking in large language modelscited by 316
- 2016Cooperative Inverse Reinforcement Learningcited by 298
- 2024Me, Myself, and AI: The Situational Awareness Dataset (SAD) for LLMscited by 78
4 more
- 2024Escalation Risks from Language Models in Military and Diplomatic Decision-Makingcited by 61
- 2024Human vs. Machine: Behavioral Differences Between Expert Humans and Language Models in Wargame Simulationscited by 20
- 2024Measuring Free-Form Decision-Making Inconsistency of Language Models in Military Crisis Simulationscited by 11
- 2025The Prompt War: How AI Decides on a Military Interventioncited by 4
Built on by
Not cited yet.
Similar, not cited
Close in wording, no citation either way
- 2026Alignment Makes Language Models Normative, Not Descriptive
- 2026Sympathetic Framing: Evaluating AI Alignment across Sociodemographic Groups
- 2025An analysis of AI Decision under Risk: Prospect theory emerges in Large Language Models
- 2026ConflictBench: Evaluating Human-AI Conflict via Interactive and Visually Grounded Environments
- 2025Large Language Models Often Know When They Are Being Evaluated
- 2026Questionnaire Responses Do not Capture the Safety of AI Agents
- 2026LLMs as Strategic Actors: Behavioral Alignment, Risk Calibration, and Argumentation Framing in Geopolitical Simulations
- 2026Principles Do Not Apply Themselves: A Hermeneutic Perspective on AI Alignment