Dual-Modality Multi-Stage Adversarial Safety Training: Robustifying Multimodal Web Agents Against Cross-Modal Attacks
Haoyu Liu, Dingcheng Li, Lukas Rutishauser, Zeyu Zheng · Mar 4, 2026 · Citations: 0
Data freshness
Extraction: FreshCheck recency before relying on this page for active eval decisions. Use stale pages as context and verify against current hub results.
Metadata refreshed
Mar 4, 2026, 6:29 PM
RecentExtraction refreshed
Mar 13, 2026, 3:01 PM
FreshExtraction source
Persisted extraction
Confidence 0.75
Abstract
Multimodal web agents that process both screenshots and accessibility trees are increasingly deployed to interact with web interfaces, yet their dual-stream architecture opens an underexplored attack surface: an adversary who injects content into the webpage DOM simultaneously corrupts both observation channels with a consistent deceptive narrative. Our vulnerability analysis on MiniWob++ reveals that attacks including a visual component far outperform text-only injections, exposing critical gaps in text-centric VLM safety training. Motivated by this finding, we propose Dual-Modality Multi-Stage Adversarial Safety Training (DMAST), a framework that formalizes the agent-attacker interaction as a two-player zero-sum Markov game and co-trains both players through a three-stage pipeline: (1) imitation learning from a strong teacher model, (2) oracle-guided supervised fine-tuning that uses a novel zero-acknowledgment strategy to instill task-focused reasoning under adversarial noise, and (3) adversarial reinforcement learning via Group Relative Policy Optimization (GRPO) self-play. On out-of-distribution tasks, DMAST substantially mitigates adversarial risks while simultaneously doubling task completion efficiency. Our approach significantly outperforms established training-based and prompt-based defenses, demonstrating genuine co-evolutionary progress and robust generalization to complex, unseen environments.