AIBearisharXiv – CS AI · 5h ago7/10
🧠
Undetectable Backdoors in Model Parameters: Hiding Sparse Secrets in High Dimensions
Researchers present Sparse Backdoor, a supply-chain attack that embeds undetectable backdoors into pre-trained image classifiers by injecting sparse perturbations masked with Gaussian noise. The attack is proven computationally infeasible to distinguish from original models under standard hardness assumptions, raising critical security concerns for AI model deployment and verification.