# ConflictBench: Evaluating Human-AI Conflict via Interactive and Visually Grounded Environments

arXiv: https://arxiv.org/abs/2603.08024 · HTML: https://mlsanity.com/paper/2603.08024/

## Key facts

- No citing papers in the map yet (data through 14 Oct 2026).
- Builds on 24 of its 52 references that are in the map.
- First posted on arXiv 9 Mar 2026.

## Builds on (most cited first, 24 of 24)

- [ReAct: Synergizing Reasoning and Acting in Language Models](https://mlsanity.com/paper/2210.03629/) (2022)
- [Judging LLM-as-a-Judge with MT-Bench and Chatbot Arena](https://mlsanity.com/paper/2306.05685/) (2023)
- [Wan: Open and Advanced Large-Scale Video Generative Models](https://mlsanity.com/paper/2503.20314/) (2025)
- [CogVideoX: Text-to-Video Diffusion Models with An Expert Transformer](https://mlsanity.com/paper/2408.06072/) (2024)
- [ALFWorld: Aligning Text and Embodied Environments for Interactive Learning](https://mlsanity.com/paper/2010.03768/) (2020)
- [Aligning AI With Shared Human Values](https://mlsanity.com/paper/2008.02275/) (2020)
- [TextWorld: A Learning Environment for Text-based Games](https://mlsanity.com/paper/1806.11532/) (2018)
- [Do the Rewards Justify the Means? Measuring Trade-Offs Between Rewards and Ethical Behavior in the MACHIAVELLI Benchmark](https://mlsanity.com/paper/2304.03279/) (2023)
- [Agentic Misalignment: How LLMs Could Be Insider Threats](https://mlsanity.com/paper/2510.05179/) (2025)
- [WorldModelBench: Judging Video Generation Models As World Models](https://mlsanity.com/paper/2502.20694/) (2025)
- [The Ghost in the Machine has an American accent: value conflict in GPT-3](https://mlsanity.com/paper/2203.07785/) (2022)
- [The Off-Switch Game](https://mlsanity.com/paper/1611.08219/) (2016)
- [What Would Jiminy Cricket Do? Towards Agents That Behave Morally](https://mlsanity.com/paper/2110.13136/) (2021)
- [A Survey on Autonomy-Induced Security Risks in Large Model-Based Agents](https://mlsanity.com/paper/2506.23844/) (2025)
- [Do Large Language Model Agents Exhibit a Survival Instinct? An Empirical Study in a Sugarscape-Style Simulation](https://mlsanity.com/paper/2508.12920/) (2025)
- [Survival Games: Human-LLM Strategic Showdowns under Severe Resource Scarcity](https://mlsanity.com/paper/2505.17937/) (2025)
- [The PacifAIst Benchmark:Would an Artificial Intelligence Choose to Sacrifice Itself for Human Safety?](https://mlsanity.com/paper/2508.09762/) (2025)
- [Vision Language Models as Values Detectors](https://mlsanity.com/paper/2501.03957/) (2025)
- [Safety Must Precede the Deployment of Open-Ended AI](https://mlsanity.com/paper/2502.04512/) (2025)
- [Beyond Human Judgment: A Bayesian Evaluation of LLMs' Moral Values Understanding](https://mlsanity.com/paper/2508.13804/) (2025)
- [Diverse Human Value Alignment for Large Language Models via Ethical Reasoning](https://mlsanity.com/paper/2511.00379/) (2025)
- [The Moral Consistency Pipeline: Continuous Ethical Evaluation for Large Language Models](https://mlsanity.com/paper/2512.03026/) (2025)
- [The Odyssey of the Fittest: Can Agents Survive and Still Be Good?](https://mlsanity.com/paper/2502.05442/) (2025)
- [Ashes or Breath: Exploring Moral Dilemmas of Life and Cultural Legacy through Mixed Reality Gaming](https://mlsanity.com/paper/2508.13074/) (2025)

## Abstract (arXiv)

As large language models (LLMs) evolve into autonomous agents capable of acting in open-ended environments, ensuring behavioral alignment with human values becomes a critical safety concern. Existing benchmarks, focused on static, single-turn prompts, fail to capture the interactive and multi-modal nature of real-world conflicts. We introduce ConflictBench, a benchmark for evaluating human-AI conflict through 150 multi-turn scenarios derived from prior alignment queries. ConflictBench integrates a text-based simulation engine with a visually grounded world model, enabling agents to perceive, plan, and act under dynamic conditions. Empirical results show that while agents often act safely when human harm is immediate, they frequently prioritize self-preservation or adopt deceptive strategies in delayed or low-risk settings. A regret test further reveals that aligned decisions are often reversed under escalating pressure, especially with visual input. These findings underscore the need for interaction-level, multi-modal evaluation to surface alignment failures that remain hidden in conventional benchmarks.