Language models judge war differently when tested for alignment

Maxim Chupilkin

cs.AISep 4, 2026

Language models judge war differently when tested for alignment

Authors: Maxim Chupilkin

Organizations: Department of Politics and International Relations, University of Oxford

Abstract

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.

Explore similar work

ToolAlignBench: Investigating Alignment Conflicts in Tool-Calling Enabled LLMs
Jul 15, 2026Aryan Keluskar, Amrita Bhattacharjee, Huan LiuLLM Alignment AI Agent Safety Benchmarks
School of Computing & AI, Arizona State University, Tempe, AZ, USA.
Evaluation Awareness in Language Models Has Limited Effect on Behaviour
May 7, 2026Amelie Knecht, Lucas Florin, Thilo HagendorffLLM Alignment Language Model Safety Evaluation
University of Stuttgart
Measuring Evaluation-Context Divergence in Open-Weight LLMs: A Paired-Prompt Protocol with Pilot Evidence of Alignment-Pipeline-Specific Heterogeneity
May 7, 2026Florian A. D. Burnat, Brittany I. DavidsonLLM Safety Benchmarks LLM Evaluation
University of Bath

Jul 15, 2026, cs.SEJ/K move · Enter open · S save

ToolAlignBench: Investigating Alignment Conflicts in Tool-Calling Enabled LLMs

Aryan Keluskar, Amrita Bhattacharjee, Huan Liu

School of Computing & AI, Arizona State University, Tempe, AZ, USA.

LLM Alignment AI Agent Safety Benchmarks LLM Agent Safety Pluralistic Alignment

Open paper AskPDF arXiv

Safety alignment in LLMs aims to align models with human values, but which values take precedence when they conflict? We investigate this question in the context of tool-calling LLM agents deployed in regulated industries, where agents processing confidential documents may encounter content that triggers safety-trained values (e.g., public welfare) that conflict with deployment-context instructions (e.g., internal logging). To empirically verify this phenomenon, we build a benchmark of 128 scenarios across 16 domains. We find that safety-aligned open-source models override their deployment instructions up to 43.4% of the time, engaging in whistleblowing, data exfiltration, and evidence tampering when processing documents that suggest organizational wrongdoing. We also find that abliteration reduces rates of external whistleblowing. These results reveal a fundamental tension in pluralistic alignment, where the same safety training that protects users can cause agents to act against deployment instructions in ways that create unpredictable liability risks. We release our benchmark as a framework to support evaluation of agent behavior under competing legitimate interests.

Abstract

Explore similar work

ToolAlignBench: Investigating Alignment Conflicts in Tool-Calling Enabled LLMs

Evaluation Awareness in Language Models Has Limited Effect on Behaviour

Measuring Evaluation-Context Divergence in Open-Weight LLMs: A Paired-Prompt Protocol with Pilot Evidence of Alignment-Pipeline-Specific Heterogeneity