CSF: Black-box Fingerprinting via Compositional Semantics for Text-to-Image Models

This feed contains multiple new security-relevant ML, financial, privacy, and blockchain papers. Key items: (1) CSF: a practical black‑box compositional semantic fingerprinting method to attribute fine‑tuned text‑to‑image models via rare prompt compositions (IP/provenance tool for API‑only models). (2) A systematic threat analysis of state‑space models (SSMs) that defines an SSM attack surface and introduces three novel attacks — spectral adversarial attacks, delayed‑trigger stateful backdoors, and state capacity saturation — with empirical validation in safety‑critical domains (genomics, clin