Network Security and Contagion

Daron Acemoğlu, Azarakhsh Malekian, Asuman Ozdaglar2013 · DOI

Abstract

We develop a theoretical model of security investments in a network of interconnected agents. Network connections introduce the possibility of cascading failures due to an exogenous or endogenous attack depending on the profile of security investments by the agents. The general presumption in the literature, based on intuitive arguments or analysis of symmetric networks, is that because security investments create positive externalities on other agents, there will be underinvestment in security. We show that this reasoning is incomplete because of a first-order economic force: security investments are also strategic substitutes. In a general (non-symmetric) network, this implies that underinvestment by some agents will encourage overinvestment by others. We demonstrate by means of examples that there can be overinvestment by some agents and also that aggregate probabilities of infection can be lower in equilibrium compared to the social optimum. We then provide sufficient conditions for underinvestment. This requires both sufficiently convex cost functions (convexity alone is not enough) and networks that are either symmetric or locally tree-like. We also characterize the impact of network structure on equilibrium and optimal investments. Finally, we show that when the attack location is endogenized (by assuming that the attacker chooses a probability distribution over the location of the attack in order to maximize damage), there is an additional incentive for overinvestment: greater investment by an agent shifts the attack to other parts of the network.

Introduction

In an interconnected world—whether we are talking about computer servers, financial institutions, or transport hubs—security is a collective problem. If your neighbor on the network fails to protect themselves, they become a vector for "contagion" that can infect you. The standard economic intuition suggests that because my security investment benefits you (a positive externality) but I don't get paid for that benefit, I will naturally invest less than what is socially ideal. This paper challenges that "underinvestment" dogma by exploring how network structure and attack types influence strategic behavior.

Methodology

The authors develop a theoretical game-theoretic model where agents (nodes) choose a security level $q$ (the probability of being immune) at a certain cost. They analyze the gap between the Nash Equilibrium (where individuals maximize their own utility) and the Social Optimum (where total network welfare is maximized).

Key technical innovations include:

  • Recursive Decomposition: A novel way to break down an agent's infection probability into an "own effect" and a "network effect" caused by others.
  • Topological Analysis: The paper moves beyond "symmetric networks" to study stars, trees, "locally tree-like" graphs, and "island" networks.
  • Attack Modeling: They compare Random Attacks (exogenous) with Strategic Attacks, where an adversary targets nodes specifically to maximize total network damage.

Key Insights: Strategic Substitutes and Network Topology

The central finding is that security investments are strategic substitutes. If my neighbor underinvests in security, the risk to me increases. This might actually compel me to invest more in my own defense to compensate for their negligence.

The Star Network Surprise

In a star network, the "center" node might invest very little, forcing the "leaf" nodes to invest heavily to protect themselves. This can result in an equilibrium where the total probability of infection is actually lower than the social optimum—the opposite of what standard theory predicts.

Figure 1: A star network with a central node (1) and peripheral leaf nodes (2-5).

Other key ideas include:

  • Sufficient Convexity: For the "underinvestment" intuition to hold, cost functions must be "sufficiently convex," meaning marginal costs must rise fast enough to prevent agents from over-responding to neighbors' risks.
  • Strategic Attacks as an "Arms Race": When an attacker targets the weakest link, agents have an additional incentive to overinvest. By beefing up my security, I am not just protecting myself from contagion; I am actively "shifting" the attacker’s gaze toward my neighbors.

Key Results

  • Trees and Underinvestment: For tree-structured networks with sufficiently convex costs, the authors prove that a unique equilibrium exists and always involves underinvestment.
  • Symmetric Networks: In symmetric networks, there is always a symmetric equilibrium that involves underinvestment. However, these same networks can also support asymmetric equilibria where some agents overinvest.
  • Strategic Attack Incentives: Under strategic attacks, the paper provides specific conditions where the "negative externality" of shifting an attack to others outweighs the positive externality of stopping contagion, leading to systemic overinvestment.

Conclusion

The research highlights that the "common wisdom" of underinvestment in network security is context-dependent. While positive externalities exist, the strategic substitutability of investments and the specific topology of the network can drive agents toward overinvestment. Future work should address dynamic contagion and endogenous network formation, where agents can choose to disconnect from high-risk nodes.

Summary updated

Abstract

We develop a theoretical model of security investments in a network of interconnected agents.Network connections introduce the possibility of cascading failures due to an exogenous or endogenous attack depending on the profile of security investments by the agents.The general presumption in the literature, based on intuitive arguments or analysis of symmetric networks, is that because security investments create positive externalities on other agents, there will be underinvestment in security.We show that this reasoning is incomplete because of a first-order economic force: security investments are also strategic substitutes.In a general (non-symmetric) network, this implies that underinvestment by some agents will encourage overinvestment by others.We demonstrate by means of examples that there can be overinvestment by some agents and also that aggregate probabilities of infection can be lower in equilibrium compared to the social optimum.We then provide sufficient conditions for underinvestment.This requires both sufficiently convex cost functions (convexity alone is not enough) and networks that are either symmetric or locally tree-like.We also characterize the impact of network structure on equilibrium and optimal investments.Finally, we show that when the attack location is endogenized (by assuming that the attacker chooses a probability distribution over the location of the attack in order to maximize damage), there is an additional incentive for overinvestment: greater investment by an agent shifts the attack to other parts of the network.
BibTeX
@misc{acemoglu2013network,
  author = {Daron Acemoğlu and Azarakhsh Malekian and Asuman Ozdaglar},
  title = {{Network Security and Contagion}},
  year = {2013},
  doi = {10.3386/w19174},
}
Figure