Privacy-Preserving Federated Learning Scheme With Mitigating Model Poisoning Attacks: Vulnerabilities and Countermeasures

Summary

This research paper shows that privacy-preserving federated learning schemes built on two honest-but-curious, non-colluding servers still leak data when defending against model poisoning. The defense computation exposes users' gradients in plaintext to one honest-but-curious server. The authors propose PBFL, a privacy-preserving, Byzantine-robust framework with a trust-scored aggregation strategy and secure subroutines over trapdoor fully homomorphic encryption, and report theoretical security analysis and experiments showing lower overhead than prior PBFL schemes.

Related items

• MediumGoogle: LLMjacking attacks rack up $1 million in cloud AI chargesSimilar attack · BleepingComputer
• LowExtracting System Prompt of LLM Services in Complex (Agentic) Settings: Hierarchical Optimization and Reliable Evaluation MetricsSimilar attack · Arxiv (cs.CR + cs.CL + cs.LG)
• InfoMCP Tool Poisoning in Agentic AI Frameworks: A Cross-Framework Measurement StudySimilar attack · Arxiv (cs.CR + cs.CL + cs.LG)
• InfoSecuring next-generation edge intelligence: A survey from federated learning to autonomous agents in cyber-physical systemsSimilar attack · OpenAlex (peer-reviewed AI security)
• MediumHow We Took Over ChatGPT AccountsSimilar attack · Simon Willison's Weblog