GhostWriter: Exploiting GPU-Cache Contention to Steal and Steer Multi-tenant Large-Language-Model Inference

Satyajit Das, Sreenath Vijayakumar. GhostWriter: Exploiting GPU-Cache Contention to Steal and Steer Multi-tenant Large-Language-Model Inference. In Chandan Karfa, Navid Asadi, Anupam Chattopadhyay, editors, Security, Privacy, and Applied Cryptography Engineering - 15th International Conference, SPACE 2025, Guwahati, India, December 16-19, 2025, Proceedings. Volume 16406 of Lecture Notes in Computer Science, pages 134-153, Springer, 2025. [doi]

Abstract

Abstract is missing.