mlsanityData to 7 Oct 2026

World model security and GUI agents

32 papers4 this weekMedium confidence in this grouping: mostly one topic, with blurry edgesMedium confidence in this grouping: mostly one topic, with blurry edges · About

Keywords: gui, attack, agents, attacks, mobile

Key facts. World model security and GUI agents: 32 papers on arXiv since 2020 (data through 7 Oct 2026). 9 of its 32 mapped papers (28%) cite “ViMo: A Generative Visual GUI World Model for App Agents” (2025, arXiv 2504.13936). 7 of its 32 mapped papers (22%) cite “MobileWorldBench: Towards Semantic World Modeling For Mobile Agents” (2025, arXiv 2512.14014). 7 new papers 17 Sep–7 Oct.

What’s new

17 Sep – 7 Oct 2026

7 new papers, 17 Sep – 7 Oct; 4 this week.

  1. This week · 4
  2. World Models Dream of Success: Diagnosing and Repairing Failure Insensitivity in Robot World Models also fits: World models for language agents
  3. AdvSim2Real : Training Web Agents Against Adaptive Prompt Injection in a Web World Model
  4. Mind the Execution Gap: Action-Semantic Mismatch in World-Model Control also fits: World models for language agents
  5. Imagine to Act: High-Fidelity Data Synthesis via Image Editing World Model for Scalable GUI Agent Training also fits: Video-generation world models and simulators
  6. Earlier · 3
  7. AutoGUIWorld: Image Generators as Visual World Models for GUI Agent
  8. The GUI Is Not the State: Diagnosing State Aliasing in GUI World Models
  9. CyberWorld: World Models for Sample-Efficient Autonomous Cyber Defense

Foundations

Most cited by papers here

  1. ViMo: A Generative Visual GUI World Model for App Agents2025cited by 9 here
  2. MobileWorldBench: Towards Semantic World Modeling For Mobile Agents2025cited by 7 here
  3. MobileDreamer: Generative Sketch World Model for GUI Agent2026cited by 7 here
  4. When World Models Dream Wrong: Physical-Conditioned Adversarial Attacks against World Models2026cited by 7 here
  5. Code2World: A GUI World Model via Renderable Code Generation2026cited by 4 here

Lineage

The papers this direction builds on, oldest first.

Each paper on the line cites the one before.

  1. 2018World Modelscited by 10 here
  2. 2018Learning Latent Dynamics for Planning from Pixelscited by 7 here
  3. 2019Dream to Control: Learning Behaviors by Latent Imaginationcited by 11 here
  4. 2020Mastering Atari with Discrete World Modelscited by 5 here
  5. 2022Transformers are Sample-Efficient World Modelscited by 4 here
  6. 2023Mastering Diverse Domains through World Modelscited by 9 here
  7. 2023Learning Interactive Real-World Simulatorscited by 4 here
  8. 2024Web Agents with World Models: Learning and Leveraging Environment Dynamics in Web Navigationcited by 11 here
  9. 2024Is Your LLM Secretly a World Model of the Internet? Model-Based Planning for Web Agentscited by 10 here
  10. 2024Understanding World or Predicting Future? A Comprehensive Survey of World Modelscited by 7 here
  11. 2025ViMo: A Generative Visual GUI World Model for App Agentscited by 9 here
  12. 2025MobileWorldBench: Towards Semantic World Modeling For Mobile Agentscited by 7 here
  13. 2026MobileDreamer: Generative Sketch World Model for GUI Agentcited by 7 here
  14. 2026Code2World: A GUI World Model via Renderable Code Generationcited by 4 here
10 more it builds on, not in this chain
  1. 2018Recurrent World Models Facilitate Policy Evolutioncited by 4 here
  2. 2022DayDreamer: World Models for Physical Robot Learningcited by 4 here
  3. 2023SafeDreamer: Safe Reinforcement Learning with World Modelscited by 4 here
  4. 2023GAIA-1: A Generative World Model for Autonomous Drivingcited by 4 here
  5. 2023TD-MPC2: Scalable, Robust World Models for Continuous Controlcited by 7 here
  6. 2024Genie: Generative Interactive Environmentscited by 6 here
  7. 2024Diffusion for World Modeling: Visual Details Matter in Ataricited by 4 here
  8. 2025Cosmos World Foundation Model Platform for Physical AIcited by 5 here
  9. 2026WebWorld: A Large-Scale World Model for Web Agent Trainingcited by 5 here
  10. 2026When World Models Dream Wrong: Physical-Conditioned Adversarial Attacks against World Modelscited by 7 here

The 24 most cited of 156.

Browse

27 of 32, ranked 6–32

  1. 6Generative Visual Code Mobile World Models20263
  2. 7TRAP: Tail-aware Ranking Attack for World-Model Planning20263
  3. 8WMAttack: Automated Attack Search for Adversarial Evaluation of World-Model Agents20263
  4. 9SafePred: A Predictive Guardrail for Computer-Using Agents via World Models20262
  5. 10How Mobile World Model Guides GUI Agents?20262
  6. 11Safety, Security, and Cognitive Risks in World Models20261
  7. 12JailWAM: Jailbreaking World Action Models in Robot Control20261
  8. 13Targeting World Models to Compromise Robot Learning Pipelines20261
  9. 14BadWorld: Adversarial Attacks on World Models20261
  10. 15Stealthy World Model Manipulation via Data Poisoning20261
All 27
  1. 16Attacking the Trusted Imagination: Oracle-Level Integrity Attacks on Imagine-then-Act World Models20261
  2. 17BadWAM: When World-Action Models Dream Right but Act Wrong20261
  3. 18SafeDream: Safety World Model for Proactive Early Jailbreak Detection20260
  4. 19PROWL: Prioritized Regret-Driven Optimization for World Model Learning20260
  5. 20ARB4WM: An Adversarial Robustness Benchmark for World Models in Continuous Control20260
  6. 21SeerGuard: A Safety Framework for Mobile GUI Agents via World Model Prediction20260
  7. 22False Prophets: On the Security of World Models in Agentic Systems20260
  8. 23Security of World-Model-Based Embodied AI: A Lifecycle of Threats, Defenses, and Evaluation20260
  9. 24DreamGuard: Efficient Runtime Guardrail for LLM Agents via Risk-Aware World Model20260
  10. 25AppDeltaWorld: Transition-Grounded Delta Code World Model for Mobile GUI Agents20260
  11. 26CIVA: Critic-Induced Value-Subspace Attacks on Visual World-Model Agents20260
  12. 27WM-R1: Training GUI Agents to Reason and leverage World Models with Reinforcement Learning20260
  13. 28GUI-CC: Benchmarking Contextual Consistency of GUI World Models as Agent Environments20260
  14. 29Solaris: Towards Interfaces That Are Generated, Not Coded20260
  15. 30TrojanWorld: Backdooring World-Model Agents via Imagination Steering20260
  16. 31When the World Lies: Backdoor Attacks on Latent World Models for Downstream Control20260
  17. 32A Critical Evaluation of Open-World Machine Learning20200