Networks that grow by "preferential attachment" follow a well-known rule: new connections tend to go to nodes that already have many connections, producing the famous "scale-free" structure where a few hubs have enormously more links than average. This pattern appears in the internet, social networks, citation graphs, and many other real systems. The question this paper asks is whether that structure is fragile: what happens when an adversary is actively trying to disrupt the network as it grows?
The researchers model an attacker who, at each step of network growth, has some probability of hijacking the process. Instead of a new edge being added in the usual preferential way, the attacker forces an existing edge to be deleted, also chosen in a preferential way (so highly connected nodes lose edges more often). The key finding is that this setup produces a sharp phase transition. When the attacker's success probability is below a critical threshold, the network remains scale-free, meaning its degree distribution still follows a power law, though with a lighter tail than an unattacked network. Once the attacker's skill crosses that threshold, the power-law property breaks down entirely and the network loses its scale-free character.
The practical upshot is that scale-free networks are neither perfectly robust nor perfectly fragile under this kind of adversarial pressure. There is a genuine tipping point. Below it, the hub-dominated architecture survives even sustained attack, which helps explain why real-world networks often seem surprisingly resilient. Above it, the architecture collapses into something qualitatively different. The paper also works out the precise mathematical form of the degree distribution in all regimes, giving a complete picture of how the structure degrades as attacker skill increases.