FedPSA: Adaptive Backdoor Attacks on Federated Large Language Models via Poisoning Sensitive Prompt Layers

Summary

FedPSA is a backdoor attack framework against federated parameter-efficient fine-tuning of LLMs. It identifies the most sensitive prompt layers, poisons a minimal subset of them under parameter constraints, and uses adaptive triggers that adjust to non-IID data. Across five datasets and six language models, it reportedly outperforms the evaluated baselines while keeping clean-task performance competitive.

Topics

• Adversarial machine learning

Related items

• MediumMalicious downloads can trick AI coding agents into compromising themselvesSimilar attack · CSO Online
• MediumGHSA-2wqp-q3c6-ccj5: MLflow: SSRF on unauthenticated MLflow servers via third-party scorer kwargs in the scorer invoke, register, and GenAI evaluate endpointsSimilar attack · GitHub Advisory Database
• MediumGHSA-vqqp-x5qm-5qp4: FastMCP MCP client wrappers construct M2M OAuth providers without issuer binding — client secret redirectable to attacker-advertised authorization serverSimilar attack · GitHub Advisory Database
• MediumNew Evidence Shows LiteLLM Secret-Stealing Workflows Ran 11 Days Before Supply Chain AttackSimilar attack · Socket Blog
• InfoVGPD-FL: Group-level poisoning detection without trusted datasets for secure aggregationSimilar attack · Elsevier Security Journals