Attention-based MIA and target label k-anonymity defence

AMIA is a new attention-based MIA tailored for tabular foudnation models that exploits attention dynamics across layers and heads to infer membership information from context set. Rather than relying on shadow models or auxiliary population data, the attack requires only access to the model. The k-anonymity-based defence targets high-risk queries, achieving a drastic reduction in attack effectiveness while preserving predictive utility.