deceptive bandit learning

A new arXiv paper (2610.09120v1) shows that an adversarial agent can exploit leaked signals merely correlated with a victim agent's randomized exploration to steer multi-agent bandit learning toward a…