paper_type.json (302B)
1 { 2 "paper_type": "empirical", 3 "reason": "Runs controlled experiments testing backdoor attacks on AI agents across multiple threat models, reports quantitative attack success rates and poisoning thresholds, and evaluates existing defense mechanisms—primary contribution is experimental findings." 4 }