Legal and Policy Considerations in Adversarial Federated Learning

A conflict exists between the adversarial vulnerabilities of federated learning and existing accountability and liability regulations in data protection and AI law. This chapter analyses GDPR, the EU AI Act, HIPAA, India's DPDP Act, and UAE ADHICS, along with technical evidence related to poisoning, backdoor, and privacy attacks. The chapter highlights how privacy-preserving architectures can reduce traceability, make causation harder to determine, and weaken single-producer product liability models. It then suggests a role-based, evidence-based framework that uses cryptographically verifiable logs, provenance records, and FL-specific contractual clauses. This framework clarifies liability, establishes safe harbors, and enables cross-border governance for high-risk federated systems.