Dynamic Watermarking and Action-Conditioned Latent Dynamics for Secure Vision-Based Controllers

Summary

This paper presents an adaptive dynamic-watermarking framework for detecting replay attacks on vision-based controllers. A private Gaussian watermark is added to the control input, and a control-conditioned latent dynamics network predicts the expected latent response, with the discrepancy from the received image serving as attack evidence. The framework is evaluated on a vision-guided UR5 manipulation task with a moving human arm, under nominal and replay-attacked observations.

Related items

• LowRailway’s AI agents go off the track on a rival’s networkSimilar attack · CSO Online
• MediumOpenAI Model Wrote Its Own Network Code to Get Around Sandbox RestrictionsSimilar attack · SecurityWeek
• MediumChainalysis Breach Exposes All Users' Credentials, Forces Shutdown of Key ServicesSimilar attack · The Hacker News
• MediumThreat actors use agentic AI tooling to find and abuse exposed secretsSimilar attack · Microsoft Security Blog
• HighCVE-2026-88980: Kubeflow Pipelines artifact fetch flaw enables server-side request forgerySimilar attack · NVD/CVE Database