W. Jiang, “PurMM: Attention-Guided Test-Time Backdoor Purification in Multimodal Large Language Models”, AAAI, vol. 40, no. 42, pp. 35562–35570, Mar. 2026.