Can an eavesdropper learn to exploit drifting quantum channels without adding noise?

The drift problem in quantum links

Quantum key distribution links are provisioned from security analyses that assume a stationary channel. The devices that determine the channel drift between recalibrations. Whether an eavesdropper who cannot alter the channel's own noise gains by following that drift has not been quantified until now.

The paper models adaptive eavesdropping as a constrained Markov decision process. The attacker selects one circuit per round. The noise level follows an Ornstein–Uhlenbeck process. The abort condition is a budget over each block of rounds. The value of adaptation is bounded by the best fixed circuit and a dynamic-programming upper bound.

How the attacker learns to adapt

The actions are learnt attacks. Earlier work by Decker et al. trained a parametrised circuit on a fixed gate template against a fixed channel. Here the gate structure and rotation angles are searched jointly. This yields circuits compact enough to form a discrete action set. The construction extends to noise models lacking a known template, including the amplitude damping channel.

The search starts from random gate sequences. It recovers the analytical cloners and the collective-attack key rate. It meets the lower bound of the Winick–Lütkenhaus–Coles objective from above.

What the numbers show for E91 and BB84

On device-independent E91 under bilateral depolarising noise, a reinforcement-learning attacker raises her Holevo information from 0.135 for the best fixed circuit to 0.348 at zero detection. That reaches 98 percent of the upper bound.

On BB84 under a drifting bit-flip channel, she exceeds a conservative noise-indexed rule by 0.024 in fidelity. That reaches 99 percent of the upper bound.

Under stationary noise, the attacker's gain from basis asymmetry changes sign between an averaged and a per-basis error-rate constraint.

Where the method reaches its limits

The work was presented as submission 202 at QCrypt 2026. A parallel work exploring the machine-learning aspects, titled "Sparsity for Free: A Budget-Induced Equilibrium in Joint Topology-Parameter Search", has been accepted for NeurIPS 2026. The paper does not say how many deployed QKD links currently operate without drift monitoring. It does not quantify the cost of adding real-time drift tracking to existing hardware.