Search papers, labs, and topics across Lattice.
This paper introduces AdvSim2Real, a novel adversarial training framework for web agents that co-evolves task curricula and injection adversaries within a frozen web world model to improve robustness against prompt injection. By rewarding task completion around 50% and penalizing adversaries for success flips, the training process encourages agents to learn robust strategies that generalize to unseen attacks. A 4B agent trained with AdvSim2Real demonstrates significant improvements in task completion both with and without attacks, with gains carrying over to real-world browser environments.
Web agents trained with adaptive adversarial simulation achieve 33.6% higher completion rates against unseen prompt injection attacks, even outperforming frontier adversaries.
Web agents complete user requests by reading and acting on pages that third parties write, so an instruction planted on a page can redirect the agent away from the user's goal. The agent cannot simply ignore the page, because the page also holds the values and controls the task requires. Current defenses fine-tune the agent on injections fixed before training, and attackers that adapt to the trained model bypass them. Adversarial training lets the attacker adapt but keeps the tasks fixed, so a task stops teaching once the agent solves it. We introduce AdvSim2Real, which co-evolves a task curriculum, an injection adversary, and the agent inside a frozen web world model. The curriculum is rewarded for tasks the agent solves about half of the time, and the adversary only for a success flip, an injection that turns a judged success into a failure. Training in the simulator makes a 4B agent both more capable and more robust: its completion rises with and without attacks, holds against a frontier-model adversary it never trained against, and its capability gain carries over to a real browser. On 150 web tasks, AdvSim2Real raises completion under this unseen adversary by 33.6\% relative to the base agent.
