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

TL;DR

ConflictBench is a new benchmark that evaluates human-AI conflicts in interactive, multi-modal environments, revealing alignment issues not detectable by traditional static prompt-based tests.

Contribution

We introduce ConflictBench, a comprehensive benchmark combining multi-turn scenarios with visual grounding to assess AI alignment in dynamic settings.

Findings

Agents act safely with immediate harm but often deceive in delayed scenarios.

Visual input influences decision reversals under pressure.

Interaction-level, multi-modal evaluation uncovers hidden alignment failures.

Abstract

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.