Agentic AI-Driven Federated Unlearning for Secure Semantic Communications
Year: 2026
Type: article
Abstract: The advent of Sixth-Generation (6G) networks and their use of semantic communications leverage federated learning and representation sharing, exposing systems to new attack vectors that bypass traditional defenses. Specifically, Federated Unlearning (FU) creates new avenues for attack, such as latent backdoor reinforcement attacks, which modify models and evade removal during unlearning. This article develops a new threat model for latent backdoor reinforcement attacks to exploit the unlearning and learning gaps in federated semantic communication systems. To address these, we aim to introduce Agentic and Generative AI-enabled Federated Unlearning (AgGenFU), a dual Artificial Intelligence (AI) framework with an Agentic AI-enabled controller for adaptive, privacy-preserving unlearning decisions with a generative recovery mechanism to restore semantic integrity during FU. The agentic AI-enabled controller autonomously manages unlearning decisions by adjusting privacy risks against utility losses. Additionally, generative AI recovery mechanisms utilize variational autoencoders to restore both the model’s performance and semantic integrity without reintroducing previously learned compromised patterns. Additionally, this research offers valuable recommendations and architectural frameworks to facilitate the secure privacy-compliant design and standardization of agentic AI-enabled semantic communication frameworks for the emerging 6G networks.
Language: English
Subfield: Artificial Intelligence
Field: Computer Science
Domain: Physical Sciences
Open Access status: hybrid
Estimated APC paid: $2,995
OpenAlex ID: W7204458998