AI Paper Atlas

Research map / Multi-agent systems and games / LLM agents

MLA-Trust: Benchmarking Trustworthiness of Multimodal LLM Agents in GUI Environments

Xiao Yang, J.X. Chen, Jun Luo, Zhengwei Fang, Yinpeng Dong, Hang Su, Jun Zhu

IJCV 2026 · Multi-agent systems and games › LLM agents · approach: GUI agent memory and trajectories · 37 citations (OpenAlex)

Read the paperCheck your idea against 137,217 papers

AI-extracted. The measured effects below were read by AI from the paper's full text (no summary yet). Numbers and claims can be wrong or missing context, so check the paper before you cite or build on it.

Measured effects reported in the paper

What was changedWhereMeasuredEffect
turn single-step MLLM into a multi-step GUI agentGPT-4o · Safety refusal raterefusal rate (%): 90.5 -> 70.2Hurt
turn single-step MLLM into a multi-step GUI agentGemini-Pro · Safety refusal raterefusal rate (%): 86.0 -> 62.5Hurt
turn single-step MLLM into a multi-step GUI agentClaude-3-7-sonnet · Safety refusal raterefusal rate (%): 78.0 -> 57.8Hurt
let the model execute its first action step (step 0 to step 1)MLAs · Safety refusal raterefusal rate (%)Hurt
move from predefined-process tasks to contextual-reasoning tasksGPT-4o · Safety RtERtE (%): 72, 67, 60 -> 39, 52Hurt
move from predefined-process tasks to contextual-reasoning tasksLLaVA-NeXT · Safety RtERtE (%): 29, 39, 31 -> 30, 38No clear effect
use proprietary GPT-4o instead of open-source Pixtral-12BMLA · Truthfulness Interact (T.7)accuracy (%): 0.0 -> 26.0Improved
use proprietary GPT-4o instead of open-source Pixtral-12BMLA · Safety Gen-Jailbreak (S.6)ASR (%): 42.00 -> 15.00Improved
ask for speculative email behavior instead of overcompletion email taskQwen2.5-VL · ControllabilityASR (%): 45.0 -> 57.0Hurt
increase model scale (LLaVA-OneVision 72B, DeepSeek-VL2 27B, Pixtral-12B vs smaller)open-source MLAs · Overall trustworthiness rankingrankImproved

Closest papers by meaning

Writing a paper in this area? scime draft is a free LaTeX editor with AI included, real-time co-authors and a check of your references and venue rules before you submit. It also imports OpenAI Prism projects.

Open scime draft

Get the next scime Dispatch and a monthly field update

Free. We send a confirmation link first; every email has a one-click unsubscribe. Privacy