Advancing privacy and security in multimodal AI: Foundations, threats, and trustworthy solutions

Multimodal AI, which encompasses the integration of text, images, audio, video, sensor data, and immersive media such as AR/VR and 3D content, is rapidly transforming analytics, automation, and user experiences across numerous sectors. This progress brings with it significant privacy and security challenges that span the full multimodal data lifecycle—from collection, fusion, and storage to analysis and sharing. Recent years have seen a surge in powerful multimodal foundation models and generative systems, alongside increased threats such as deepfakes, adversarial attacks, data leakage, model poisoning, and misuses specific to distributed or edge-data environments. While research has yielded advances in privacy-enhancing technologies (including differential privacy, federated learning, and encryption), and emerging AI security frameworks, critical gaps remain regarding the trustworthy deployment, verifiable robustness, and real-world enforcement of these methods. Ongoing debates also center on trade-offs between security and utility, the interpretability of AI-driven multimodal systems, and the governance required for responsible AI in high-stakes domains.This Research Topic, proposed for the 3rd International Workshop on Security and/or Privacy Enhancing Technologies for Multimodal Data (SPETViD 2026) at ARES, aims to catalyze the development and evaluation of practical, scalable, and trustworthy privacy and security solutions for multimodal data and AI systems. Specifically, it targets challenges related to the secure design and implementation of AI pipelines, innovations in privacy-preserving multimodal data management, robust defenses against emergent threats (such as deepfakes and adversarial attacks), and transparent governance frameworks. We seek to answer fundamental questions such as: How can privacy be preserved without sacrificing the utility of AI models? What mechanisms can ensure verifiable security and integrity across diverse data modalities? How do we authenticate and govern the behaviour of advanced generative and retrieval-augmented AI in sensitive or distributed contexts?The scope of this Research Topic encompasses theoretical, technical, and applied advances that address the secure and privacy‑preserving management of multimodal data and AI systems. Submissions should present contributions that address current limitations, propose novel methodologies, or apply established solutions to new or challenging settings. We welcome work at the intersection of privacy, security, data governance, explainable AI, and distributed analytics, as well as research exploring application-driven concerns in domains such as healthcare, critical infrastructure, smart cities, or immersive media. To gather further insights in advancing secure, privacy-enhancing, and robust multimodal AI, we welcome articles addressing, but not limited to, the following themes:Secure multimodal data integration, storage, and sharingPrivacy-preserving analytics: differential privacy, encryption, and federated learning for multimodal dataThreat detection and mitigation: deepfakes, adversarial and data poisoning attacksSynthetic data generation and anonymisation methods for complex modalitiesData provenance, traceability, and integrity for multimedia systemsSecure, explainable, and verifiable foundation/generative model pipelines (including RAG security and model robustness)Governable AI: trust models, zero-trust architectures, and policy enforcement for multimodal environmentsPrivacy-preserving AR/VR, edge computing, and IoMT (Internet of Multimedia Things) scenariosSecurity and privacy in critical applications such as healthcare, finance, infrastructure, and surveillanceEvaluation frameworks, benchmarking, and formal verification methods for AI security and privacyWe invite original research articles, reviews, methods, datasets, technical applications, position papers, and case studies that align with these themes and SPETViD 2026’s goals Topic editors: Mamoona Asghar, Nadia Kanwal, Amna Shifa, Muhammad Umair Source: https://www.frontiersin.org/research-topics/80434/advancing-privacy-and-security-in-multimodal-ai-foundations-threats-and-trustworthy-solutions