Call for Papers Open for CVPR 2026 Workshop on Multimodal Agent Security

AI is advancing fast, and the need to keep the systems being built secure and reliable is growing with it. The question is especially pressing for multimodal large language models, which are becoming more powerful and more autonomous. The sixth AdvML@CV workshop, devoted entirely to this topic, will take place at CVPR 2026.

Multimodal large language models (LLMs) are systems that can process and generate information in several formats: text, images, audio and video. When built into agents, they can make decisions and act in the real world. That opens up huge opportunities but also creates new risks. Imagine an LLM agent running a complex production line or an autonomous vehicle. In the event of a failure or a targeted attack, the consequences could be catastrophic.

AdvML@CV aims to bring together researchers and practitioners working on machine learning security to share experience and develop strategies for defending against potential threats. The workshop will pay particular attention to the following topics: attacks on multimodal LLMs, methods for detecting and preventing such attacks, making models more robust to adversarial inputs, and developing secure architectures and protocols.

Taking part in AdvML@CV is a chance to learn about the latest advances in multimodal LLM security, present your own work and connect with leading experts. Submissions are already open, and the organizers invite all interested parties to present their research and developments.

Multimodal LLM security is not merely an academic problem but a pressing necessity. The future of many industries and the well-being of society as a whole depend on how well we can protect these systems from malicious actors and keep them running reliably. AdvML@CV is an important venue for discussing these issues and working out joint solutions. Taking part will not only broaden your knowledge but also help build a safer and more reliable AI world.

In short, AdvML@CV at CVPR 2026 is a key event for anyone interested in the security of multimodal large language models. It is a chance to learn about the latest research, share experience and contribute to the reliability of these powerful technologies. Don't miss the chance to join this important conversation.