How does Random Forest behave under adversarial feature correlation injection?

Adversarially introduced correlated features can reduce Random Forest effectiveness by increasing tree similarity.

If adversarial features are engineered to be highly correlated with strong predictors, feature subsampling may still repeatedly select them across trees. This increases covariance among trees and reduces ensemble diversity. Consequently, variance reduction collapses and RF behaves closer to a single deep tree ensemble with limited gain.