Membership Inference Attacks for Retrieval Based In-Context Learning for Document Question Answering

Collection of new security-research preprints (arXiv) describing multiple practical risks and mitigation directions: (1) Privacy: two black-box membership‑inference attacks against retrieval-augmented in‑context learning that work with paraphrased queries and a lightweight ensemble prompting defense that reduces leakage. (2) Post‑quantum transition: a systematization of post‑quantum (PQ) network architectures, threat models (including harvest‑now, decrypt‑later), and trade-offs for PQ‑PKI vs symmetric or hybrid approaches. (3) Model supply‑chain/backdoors: Sparse Backdoor — a provably undetect