Danger actors can target AI products for theft, reverse engineering or unauthorized manipulation. Attackers could compromise a product’s integrity by tampering with its architecture, weights or parameters; the core factors that ascertain a model’s actions, precision and functionality.Variational autoencoders or VAEs, which were introduced in 2