General Sum Frameworks for Strategic Security Games

You are reading version 1 of this direction, frozen 2026-10-04. The latest version is version 2.

Version 1 of 264 papers67 conceptsfrozen 2026-10-04

In traditional game theory, security is often modeled as a "zero-sum" interaction: every bit of ground a defender gains is exactly what an attacker loses. This assumption is mathematically elegant because it allows researchers to use efficient linear programming to find optimal strategies. However, as recent research across 60+ papers suggests, the zero-sum model is often a "convenient fiction" that fails to capture the messy reality of cyber-physical systems, conservation, and urban defense.

The emerging research direction focuses on General-Sum Security Games, where the payoffs for the defender and attacker are asymmetric. This shift allows models to account for operational costs, interdependent risks, and the "price of usability" in human-centric security.

The Failure of the Zero-Sum Assumption

The core intuition behind the shift to general-sum models is that the motivations of an attacker rarely mirror the losses of a defender. As Kien C. Nguyen and Tansu Alpcan argue in Stochastic Games for Security in Networks with Interdependent Nodes, a defender might suffer massive operational losses from a data breach, while the attacker’s primary gain is the specific value of the exfiltrated data. These are not perfectly inverse.

Furthermore, defense is never free. Mário S. Alvim points out in Information Leakage Games that deploying security measures introduces overhead—such as increased latency or power consumption—that penalizes the defender without providing any benefit to the attacker. In a zero-sum world, these "mutual losses" are impossible to represent.

Modeling Interdependence and Cascades

Real-world security isn't just about protecting a single point; it’s about managing a network where nodes are linked. Kien C. Nguyen introduced frameworks that combine stochastic games with linear influence networks. This allows defenders to model the "domino effect": if one server is compromised, the security value and vulnerability of its neighbors shift dynamically.

This interdependence extends to environmental protection. In Efficient Resource Allocation for Protecting Coral Reef Ecosystems, Bo An and Yue Yin demonstrate that the ecological loss of a destroyed reef is far greater than the black-market profit a poacher might gain. By treating this as a general-sum game, they can more accurately allocate patrol boats to areas where the "value-at-risk" is highest, rather than just where the attacker's gain is greatest.

Scaling to the Real World

Moving away from zero-sum assumptions increases computational complexity. To solve these general-sum interactions at scale, researchers have developed several key mechanisms:

  1. Marginal Probabilities: Instead of calculating every possible joint distribution of security checkpoints, the RANGER framework by Jason Tsai and Milind Tambe uses polynomial-sized linear programs to focus on the probability of a single resource being at a specific location.
  2. Double Oracle Algorithms: To handle massive "action spaces" (like all possible patrol routes in a forest), researchers use the Double Oracle approach. As seen in the work of Yi Wu on Green Security Games, the system starts with a small set of strategies and iteratively adds the "best response" for both the attacker and defender until an equilibrium is reached.
  3. Deep Reinforcement Learning: For continuous environments, Nitin Kamra and Fei Fang developed OptGradFP, which uses neural networks to learn optimal policies offline, allowing for millisecond-level decision-making in the field.

The Human Element: Operationalizability

A mathematically "optimal" strategy is useless if a human guard cannot follow it. Sara Marie McCarthy and Arunesh Sinha identified the Price of Usability in The Price of Usability: Designing Operationalizable Strategies for Security Games. They argue that security models must limit the complexity of protocols to what personnel can actually implement. By constraining the game to a smaller set of "operationalizable" actions, agencies can achieve near-optimal protection while reducing human error and fatigue.

From Conflict to Persuasion

Perhaps the most counter-intuitive result of general-sum modeling is the use of strategic deception. Muhammed O. Sayin and Tamer Basar propose in their Deception-As-Defense Framework that because attacker and defender interests are not perfectly opposed, a defender can "craft" the information an adversary sees. By guiding an attacker toward a specific goal that is less damaging to the system, the defender can effectively "persuade" the adversary to fulfill a benign task while they believe they are succeeding in their attack.

This evolution from simple zero-sum competition to complex, general-sum strategic interaction represents a fundamental shift in how we secure everything from airport terminals to endangered species.

Papers in this version

Showing 60 of 64 papers.

History & citationVersion 2 · 2026-10-04

Current · Permanent link

  1. Version 2 (latest) · 2026-10-04 · 40 papers
  2. Version 1 · 2026-10-04 · 64 papers
Cite this direction

Latest version (a living page):

@misc{lacuna_zero_sum_game_assumption_63e1101f,
  title = {zero-sum game assumption},
  author = {{Lacuna}},
  howpublished = {Lacuna research direction},
  year = {2026},
  url = {https://lacuna.tiptreesystems.com/direction/zero-sum-game-assumption/rdr_63e1101f4d6249ff8138588fc8b66436},
  note = {Latest version of a living research direction; cite a version for a fixed text. Accessed 2026-11-20},
}

This exact version (version 1, frozen text):

@misc{lacuna_general_sum_frameworks_for_strategic_sec_63e1101f_v1,
  title = {General Sum Frameworks for Strategic Security Games},
  author = {{Lacuna}},
  howpublished = {Lacuna research direction},
  year = {2026},
  url = {https://lacuna.tiptreesystems.com/direction/general-sum-frameworks-for-strategic-security-games/rdr_63e1101f4d6249ff8138588fc8b66436/v/1},
  note = {Version 1, frozen 2026-10-04. Accessed 2026-11-20},
}