Abstract

Despite the swift adoption of FL within distributed cloud systems, boosting collaborative intelligence and maintaining data locality, current FL frameworks are susceptible to a variety of malicious threats, including adversarial bot intrusions, data poisoning, privacy breaches, and new quantum-based cryptographic attacks. Existing CAPTCHA technologies are also shown to have poor resistance to AI-driven solvers, with accuracy reaching 78.4% against complex distributed system-based attacks. Accordingly, this paper introduces an AI-Optimized Quantum-Resistant CAPTCHA-based Secure Computation Framework (AQC-SCF) for privacy-preserving FL within distributed cloud environments. The proposed framework combines an adaptive adversarial CAPTCHA generation engine, post-quantum lattice-based cryptography, and a secure aggregation scheme to enhance authentication, data privacy, and model integrity in a collaborative learning environment. Evaluated the performance of the proposed scheme across 120 distributed clients running on a simulated cloud, over 50 communication rounds, using more than 85,000 training samples from benchmark classification datasets. The experiments indicate that compared to standard FL, secure FL, and blockchain-based FL, the proposed AQC-SCF achieves 97.8%, 96.9%, 96.3%, and 96.6% accuracy, precision, recall, and F1-score, respectively, which improves standard FL, secure FL, blockchain-based FL, and AQC-FL models by 8.7%, 6.1%, 4.4%, and 2.6%. Authentication breaches were reduced by 93.5%, and communication overhead decreased by 27.8%, while quantum-attack-based cryptographic resilience increased by 41.6% with an average latency of 0.84s per aggregation round. These findings demonstrate the suitability and scalability of the proposed scheme for securely realizing distributed federated intelligence in next-generation cloud environments.