SecureGate: Learning When to Reveal PII Safely via Token-Gated Dual-Adapters for Federated LLMs
Accepted and published in ACL 2026, 2026.
We study when and how a federated LLM should expose potentially sensitive information, using a dual-adapter design and token-level gating to separate global from sensitive knowledge while keeping inference overhead practical.
Full paper: arXiv:2602.13529.