Asymmetric-Information Resource Allocation Games: An LP Approach to Purposeful Deception

A team from Georgia Tech (Longxu Pan, Yue Guan, Daigo Shishika, Panagiotis Tsiotras) has published a paper on arXiv introducing the Deceptive Resource Allocation Game (DRAG), a Bayesian game framework that models purposeful deception in resource allocation. In DRAG, a Defender allocates limited resources across a true asset and several decoys to influence an Attacker's beliefs and actions, aiming to divert the Attacker away from the real target. The key innovation is characterizing "purposeful deception"—where the Defender only deceives when it improves overall performance, not as a default behavior.

The researchers solve for the Perfect Bayesian Nash Equilibrium (PBNE) of this asymmetric-information game, showing that despite the complex interdependence between beliefs and policies, the problem admits an efficient, non-iterative linear programming formulation. Numerical results demonstrate that the resulting policies naturally balance effective resource allocation with belief manipulation, giving rise to emergent deceptive behaviors that are purposeful rather than arbitrary. This work bridges game theory and multi-agent systems, offering a mathematically rigorous framework for understanding when and how AI systems should employ deception strategically. The paper is available on arXiv (2604.25070) and has applications in cybersecurity, military strategy, and autonomous systems where protecting critical assets is paramount.