# Adversarial co-learning exposes bottleneck fragility in quantum repeater networks

_Monday, July 13, 2026 at 4:07 AM EDT · Science, AI · Latest · Tier 2 — Notable_

Researchers modeled entanglement-based quantum routing as an adversarial bandit game: Alice selects an end-to-end repeater path for an Ekert-91 protocol while an adversary, Eve, chooses an attack surface, either edge intercept-resend or repeater memory degradation. Payoffs come from cached SeQUeNCe simulations, and Alice accepts a turn when a finite-sample statistic violates the CHSH bound. Across 50 structured topologies, learned edge-retention strategies track a full-matrix minimax reference closely (Pearson r = 0.99). Under a single-surface attack model, bottleneck edge families receive zero retention while non-bottleneck families follow a 1 − 1/N coverage rule. Decision-tree explanations fitted to graph-, attack-, and route-level targets achieve high faithfulness, and the authors package the trees into prompts for local language models to generate natural-language summaries, an open-source explanation pipeline for quantum-network game outcomes.

## Sources

- [arXiv](https://arxiv.org/abs/2607.09378)

---
Canonical: https://techandbusiness.org/newswire/A9KxU337ELsycETkoq1u7h
Retrieved: 2026-07-13T10:49:46.138Z
Publisher: Tech & Business (techandbusiness.org)