SeerGuard: A Safety Framework for Mobile GUI Agents via World Model Prediction

arXiv 2026 33.4 method, system, application

TLDR

SeerGuard uses world model prediction to proactively assess risks before executing actions in mobile GUI agents, improving safety.

Reasoning

The paper proposes a novel proactive safety framework using a safety-augmented world model, showing strong empirical gains on a specific agent. However, the evaluation is limited to one agent type and does not discuss generalization to other GUI environments or real-world deployment challenges.

Read-first score

Read-first score 33.4, weighted from topical fit, citation, graph, method, reproducibility, and recency signals. Original total remains 32.

Recency 6%

100

Uses a gentle age decay so recent papers surface without erasing older foundations. 2026

Methodology quality 18%

Screens visible abstract and analysis fields for experiment, dataset, baseline, metric, and limitation evidence. markers=experiment

Topical relevance 29%

45.7

Uses existing LLM keyword relevance scores normalized to 0-100. world model,world simulator,generative world model,interactive world model,video world model,world dynamics prediction,model-based reinforcement learning world model

Reproducibility 18%

Screens links and visible text for paper, code, dataset, artifact, and repository signals. pdf=True; code=False; dataset=False; markers=none

Citation impact 18%

Uses OpenAlex-shaped citation metadata as a bibliometric attention signal, separate from paper quality. cited_by_count=0

Citation velocity 12%

Citation velocity estimates citations per publication-year to reduce old-paper bias. velocity=0.00

Field roles

Frontier

Rank sensitivity

Stability: volatile; rank range: 138.

Keyword Scores

world model

world dynamics prediction

model-based reinforcement learning world model

world simulator

interactive world model

generative world model

video world model

Deep Analysis

Innovations

Pre-execution instruction-level screening and action-level risk assessment for mobile GUI agents
Consequence-aware action-level assessment that anticipates likely outcomes before execution
Unified Safety-Augmented World Model (SAWM) via multi-task learning integrating semantic next-state prediction and safety risk assessment

Methodology

SeerGuard uses pre-execution instruction-level screening and action-level risk assessment, where the action-level assessment analyzes agent-proposed actions within current GUI states to predict likely outcomes and identify risks. The framework is built on a unified SAWM trained via multi-task learning to jointly perform semantic next-state prediction and safety risk assessment. It is evaluated on diverse mobile GUI agents, measuring safety-utility and risk-cost scores.

Key Results

On Qwen3-VL-8B-Instruct, SeerGuard increases the safety-utility score from 0.191 to 0.596 at ω=0.8 and reduces the risk-cost score from 0.347 to 0.130 at α=0.8. The SAWM validates instruction-level screening and action risk assessment/next-state prediction capabilities.

Resources

Paper