No-go theorem for device-independent security in relativistic causal theories

A fundamental question in device-independent quantum cryptography is to determine the minimal physical principle upon which the security of such a cryptographic protocol (such as for key distribution or for randomness generation) may be based. Since the seminal work on device-independent quantum key distribution by J. Barrett, L. Hardy, and A. Kent [Phys. Rev. Lett. 95, 010503 (2005)PRLTAO0031-900710.1103/PhysRevLett.95.010503], a conjectured candidate for certification of device-independent security has been the principle of relativistic causality, namely the disallowance of causal loops. While this principle has thus far been equated with the no-signaling constraints, it has been shown recently that in multiparty Bell scenarios, the no-signaling constraints are sufficient but not necessary for relativistic causality, and a refined set of constraints has been proposed that more precisely capture the notion of relativistic causality. In this paper, we build upon this finding to show that, in contrast to the no-signaling constraints, the constraints of relativistic causality are not sufficient for certification of device-independent security. More specifically, we show that there exist correlations allowed by the causality principle that allow an adversary to gain complete information about the measurement outcomes of honest parties in any device-independent cryptographic protocol, thereby rendering the protocol completely insecure. As a tool to develop this adversarial attack strategy, we fully characterize the set of correlations allowed by relativistic causality in the tripartite Bell scenario of three parties, each performing two binary measurements, that may be of independent interest. We also demonstrate the difference between the relativistic causal correlations and those allowed by the usual no-signaling constraints by presenting explicit communication tasks wherein the two sets exhibit striking difference in their respective winning probabilities.