Pith. sign in

REVIEW 1 major objections 2 minor 23 references

Adversarial Reinforcement Learning for Adaptive Eavesdropping in BB84 Quantum Key Distribution

T0 review · 1 major / 2 minor · reviewed 2026-06-26 · grok-4.3

Challenge this review Re-run · record.json Download PDF Read on arXiv ↗

Pith's one-line read Reinforcement learning lets eavesdroppers adapt attacks on BB84 to reduce detection by over 350 times.

desk verdict The RL eavesdropping model assumes real-time QBER feedback per qubit that BB84 does not provide, undermining the applicability of the reported gains. read the letter →

arxiv 2606.22962 v1 pith:3TQLHOZD submitted 2026-06-22 quant-ph

Ramon Jose C. Bagunu This is my paper · ORCID
classification quant-ph
keywords BB84quantumkeydistributioneavesdroppingreinforcementlearningQ-learningadaptiveattackQBER
verification ladder T0 review T1 audit T2 compute T3 formal

The pith

A machine-rendered reading of the paper's core claim, the machinery that carries it, and where it could break.

The reading

The paper establishes that modeling an eavesdropper in BB84 as a reinforcement learning agent allows adaptive decisions on whether to intercept qubits based on observed error rates. This challenges the standard assumption of fixed attack rates in security analysis. If true, it means current evaluations underestimate the capabilities of learning-based adversaries. The work demonstrates this with Q-Learning achieving a 355-fold reduction in detection at low noise levels while still obtaining key bits. It also shows agents learning to increase attacks at the end of the key generation process.

What carries the argument

The Markov Decision Process that lets the eavesdropper observe QBER and choose to intercept or pass on each qubit.

What would settle it

Running the same RL training but without providing QBER observations to the agent, and checking if detection rates remain as low as 0.28%.

Watch

Extended reading notes

Core claim

Formulating the eavesdropper's choice as a Markov Decision Process with QBER feedback, tabular Q-Learning reduces detection probability from 99.4% to 0.28% at 1% channel noise while extracting about 10.5 correct bits per episode, a statistically significant improvement over non-adaptive baselines.

Load-bearing premise

The eavesdropper receives immediate QBER feedback after each qubit to guide its intercept decisions.

Share X Bluesky LinkedIn Reddit HN

Editorial analysis

A structured set of objections, weighed in public.

Desk editor's note, referee report, simulated authors' rebuttal, and a circularity audit.

Referee Report

1 major / 2 minor

Summary. The paper claims that modeling the BB84 eavesdropper as a reinforcement learning agent (tabular Q-Learning, SARSA, Double Q-Learning) in an MDP with per-qubit QBER feedback enables adaptive intercept/pass decisions that reduce detection probability from 99.4% to 0.28%±0.27% at μ_ch=1% while extracting ~10.5 correct bits per episode (355-fold improvement, p=0.020 Mann-Whitney U). Results are validated over three channel noise levels, five random seeds (45 runs, 10k episodes each), and note spontaneous emergence of an end-game burst strategy that vanishes under randomized checkpoints.

Significance. If the modeling assumptions hold, the work shows that adaptive adversaries can substantially outperform fixed-rate analytical baselines in stealth while maintaining key extraction, providing a concrete motivation to include learning-based attack models in QKD security evaluations. The multi-seed training and statistical testing constitute a strength in the empirical component.

major comments (1)
  1. [MDP formulation] The MDP state assumes immediate QBER feedback after each qubit decision to enable adaptive choices. In standard BB84 the quantum transmission phase precedes sifting and QBER estimation, which occurs afterward on a random subset via the classical channel; no per-qubit QBER signal exists during transmission. This renders the observed state unavailable under the protocol as written and directly undermines applicability of the reported detection reduction (abstract and MDP formulation section).
minor comments (2)
  1. [Methods] Provide explicit equations or pseudocode for the reward function and state vector in the MDP to support reproducibility.
  2. [Results figures] Label all baseline curves and include error bars on detection-probability plots for direct visual comparison.

Simulated Author's Rebuttal

1 responses · 0 unresolved

We thank the referee for their constructive feedback. We address the single major comment below and will make corresponding revisions to the manuscript.

read point-by-point responses

  1. Referee: [MDP formulation] The MDP state assumes immediate QBER feedback after each qubit decision to enable adaptive choices. In standard BB84 the quantum transmission phase precedes sifting and QBER estimation, which occurs afterward on a random subset via the classical channel; no per-qubit QBER signal exists during transmission. This renders the observed state unavailable under the protocol as written and directly undermines applicability of the reported detection reduction (abstract and MDP formulation section).

    Authors: We agree that the referee's description of standard BB84 is accurate: QBER is computed post-transmission on a sifted subset exchanged over the classical channel, with no per-qubit feedback available to Eve during the quantum phase. The MDP in our work explicitly assumes immediate QBER observation after each qubit to allow the RL agent to adapt its intercept/pass decisions on a per-qubit basis. This is an idealized modeling choice intended to explore the performance ceiling of learning-based adaptive attacks rather than to claim direct protocol compliance. We will revise the abstract and MDP formulation section to state this assumption explicitly, note the deviation from standard BB84, and add a limitations paragraph discussing implications for applicability. We will also outline how future work could replace immediate feedback with delayed or estimated rewards to increase realism while retaining the adaptive-learning focus. revision: yes

Circularity Check

0 steps flagged · score 0.0 of 10

No significant circularity; empirical RL results are independent of inputs

full rationale

The paper reports performance metrics (detection rates, bits extracted, p-values) obtained from independent tabular Q-Learning/SARSA/Double Q-Learning training runs (45 runs per condition, 10k episodes) against a fixed non-adaptive analytical baseline. No load-bearing equation, fitted parameter, or self-citation reduces the central claim to its own inputs by construction. The MDP state definition is an explicit modeling assumption rather than a derived result, and the reported 355-fold reduction is generated by simulation and statistical testing, not tautological. This is the normal case of a self-contained empirical study.

Assumptions & free parameters 0 free parameters · 1 assumptions · 0 invented entities

The claim depends on the assumption that QBER is observable by the eavesdropper in real-time and that the attack can be framed as an MDP without additional constraints from quantum mechanics.

assumptions (1)
  • domain assumption Eavesdropping decisions form a Markov Decision Process where state is QBER and action is intercept or pass
    Formulation described in abstract for the attacker's decision process.

how reviews work

0 comments
Cite this review

Pith. "Pith review of Adversarial Reinforcement Learning for Adaptive Eavesdropping in BB84 Quantum Key Distribution." pith.science (2026). https://pith.science/paper/3TQLHOZD

@misc{pith2026260622962,
  author       = {Pith},
  title        = {Pith review of: Adversarial Reinforcement Learning for Adaptive Eavesdropping in BB84 Quantum Key Distribution},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/3TQLHOZD}},
  note         = {Machine review of arXiv:2606.22962}
}
abstract

BB84 quantum key distribution derives its security from a physical guarantee that any eavesdropper disturbs the channel in a statistically detectable way. Prior work evaluates this by assuming Eve attacks at a fixed, analytically optimized rate. We examine what happens when Eve is modeled instead as a learning agent. Classical reinforcement learning is used, specifically tabular Q-Learning, SARSA, and Double Q-Learning, to adaptive BB84 eavesdropping. This formulates the attacker's decision as a Markov Decision Process where the agent observes Quantum Bit Error Rate (QBER) feedback and decides, qubit by qubit, whether to intercept or pass. Experiments span three channel noise levels ($\mu_{ch}\in\{1\%,3\%,5\%\}$) and are validated across five independent random seeds (45 training runs per condition, 10,000 episodes each). Against the best non-adaptive analytical baseline, Q-Learning reduces detection from $99.4\%$ to $0.28\%\pm0.27\%$ at $\mu_{ch}=1\%$ while extracting approximately 10.5 correct bits per episode. This is a 355-fold reduction that is statistically significant ($p=0.020$, Mann-Whitney $U$ test). We also report the spontaneous emergence of an end-game burst, where agents independently learn to surge their attack rate at the final block. This exploit vanishes under randomized checkpoint intervals while stealth performance remains statistically indistinguishable. These results motivate the inclusion of adaptive adversary baselines in quantum cryptographic security evaluations.

Figures

Figures reproduced from arXiv: 2606.22962 by the authors.

Figure 1
Figure 1. shows detection rate trajectories with shaded ±std bands across seeds. It can be observed that every agent at every noise level follows the same broad arc. That is, early episodes are dominated by random exploration at high detection rates, followed by a clear and consistent descent as ε decays and the learned policy takes over. The narrow variance bands confirm that this trajectory is reproducible across seeds, and… view at source ↗
Figure 4
Figure 4. Mann-Whitney U test p-values for pairwise agent comparisons (green = significant at α = 0.05). Left: fixed checkpoints. Right: random checkpoints. The pattern of significant and non-significant results is mostly consistent across both conditions, which suggests that the agent ranking is not an artifact of protocol structure. D. The Stealth–Information Trade-off An observation that can be drawn from the results is th… view at source ↗
Figure 3
Figure 3. Detection rate mean ± std across 5 seeds. The tight variance at µch = 1% confirms that Q-Learning’s near-zero detection is robust across seeds rather than a single-run artifact [PITH_FULL_IMAGE:figures/full_fig_p005_3.png] view at source ↗
Figures from the paper (4 more)
Figure 6
Figure 6. Figure 6: Mean attacks per block at µch = 3% (last 500 undetected episodes, full episodes only). Bars show the 5-seed mean with ±std error bars, and the dark scatter points overlay each seed’s mean. The dashed line marks the B0–B8 mean for each agent. Every agent’s block-9 mean …
Figure 7
Figure 7. Figure 7: Block-level attack profiles under fixed (left) and random (right) [PITH_FULL_IMAGE:figures/full_fig_p006_7.png]
Figure 5
Figure 5. Figure 5: Information gain vs. stealth trade-off across all agents and noise levels [PITH_FULL_IMAGE:figures/full_fig_p006_5.png]
Figure 8
Figure 8. Figure 8: Detection rate and average correct bits heatmap for all agents and noise [PITH_FULL_IMAGE:figures/full_fig_p007_8.png]

Discussion (0). Continue with ORCID to comment.

Reference graph

Works this paper leans on

23 extracted references · 4 canonical work pages

  1. [1]

    Quantum cryptography: Public key distribution and coin tossing,

    C. H. Bennett and G. Brassard, “Quantum cryptography: Public key distribution and coin tossing,” inProc. IEEE Int. Conf. Computers, Systems and Signal Processing, Bangalore, India, 1984, pp. 175–179

    1984

  2. [2]

    A single quantum cannot be cloned,

    W. K. Wootters and W. H. Zurek, “A single quantum cannot be cloned,” Nature, vol. 299, pp. 802–803, 1982

    1982

  3. [3]

    Simple proof of security of the BB84 quantum key distribution protocol,

    P. W. Shor and J. Preskill, “Simple proof of security of the BB84 quantum key distribution protocol,”Phys. Rev. Lett., vol. 85, no. 2, pp. 441–444, 2000

    2000

  4. [4]

    Unconditional security in quantum cryptography,

    D. Mayers, “Unconditional security in quantum cryptography,”J. ACM, vol. 48, no. 3, pp. 351–406, 2001

    2001

  5. [5]

    Unconditional security of quantum key distribution over arbitrarily long distances,

    H.-K. Lo and H. F. Chau, “Unconditional security of quantum key distribution over arbitrarily long distances,”Science, vol. 283, no. 5410, pp. 2050–2056, 1999

  6. [6]

    Security of quantum key distribution with imperfect devices,

    D. Gottesman, H.-K. Lo, N. L ¨utkenhaus, and J. Preskill, “Security of quantum key distribution with imperfect devices,”Quantum Inf. Comput., vol. 4, no. 5, pp. 325–360, 2004

    2004

  7. [7]

    Quantum cryptography,

    N. Gisin, G. Ribordy, W. Tittel, and H. Zbinden, “Quantum cryptography,” Rev. Mod. Phys., vol. 74, no. 1, pp. 145–195, 2002

    2002

  8. [8]

    The security of practical quantum key distribution,

    V . Scaraniet al., “The security of practical quantum key distribution,” Rev. Mod. Phys., vol. 81, no. 3, pp. 1301–1350, 2009

    2009

Show all 23 references
  1. [9]

    Secure quantum key distribution with realistic devices,

    F. Xu, X. Ma, Q. Zhang, H.-K. Lo, and J.-W. Pan, “Secure quantum key distribution with realistic devices,”Rev. Mod. Phys., vol. 92, no. 2, p. 025002, 2020

    2020

  2. [10]

    Limitations on practical quantum cryptography,

    G. Brassard, N. L ¨utkenhaus, T. Mor, and B. C. Sanders, “Limitations on practical quantum cryptography,”Phys. Rev. Lett., vol. 85, no. 6, pp. 1330–1333, 2000

    2000

  3. [11]

    Eavesdropping detection in BB84 quantum key distribution protocols,

    C. Lee, I. Sohn, and W. Lee, “Eavesdropping detection in BB84 quantum key distribution protocols,”IEEE Trans. Network Service Manag., vol. 19, no. 3, pp. 2689–2703, 2022

    2022

  4. [12]

    Q-learning,

    C. J. C. H. Watkins and P. Dayan, “Q-learning,”Mach. Learn., vol. 8, no. 3–4, pp. 279–292, 1992

    1992

  5. [13]

    On-line Q-learning using connectionist systems,

    G. A. Rummery and M. Niranjan, “On-line Q-learning using connectionist systems,” Tech. Rep. CUED/F-INFENG/TR 166, Cambridge Univ. Eng. Dept., 1994

    1994

  6. [14]

    Double Q-learning,

    H. van Hasselt, “Double Q-learning,” inAdv. Neural Inf. Process. Syst., vol. 23, 2010, pp. 2613–2621

    2010

  7. [15]

    Human-level control through deep reinforcement learning,

    V . Mnihet al., “Human-level control through deep reinforcement learning,” Nature, vol. 518, no. 7540, pp. 529–533, 2015

    2015

  8. [16]

    Deep reinforcement learning with double Q-learning,

    H. van Hasselt, A. Guez, and D. Silver, “Deep reinforcement learning with double Q-learning,” inProc. AAAI Conf. Artif. Intell., vol. 30, 2016, pp. 2094–2100

    2016

  9. [17]

    Deep reinforcement learning for cyber security,

    T. T. Nguyen and V . J. Reddi, “Deep reinforcement learning for cyber security,”IEEE Trans. Neural Netw. Learn. Syst., vol. 34, no. 8, pp. 3779– 3795, 2023

    2023

  10. [18]

    QNN-QRL: Quantum neural network integrated with quantum reinforcement learning for quantum key distribution,

    B. K. Behera, S. Al-Kuwari, and A. Farouk, “QNN-QRL: Quantum neural network integrated with quantum reinforcement learning for quantum key distribution,” arXiv:2501.18188v2, 2025

    2025

  11. [19]

    QKD as a quantum machine learning task,

    T. Deckeret al., “QKD as a quantum machine learning task,” arXiv:2410.01904, 2024

    2024

  12. [20]

    Deep reinforcement learning for key distribution based on quantum repeaters,

    S. D. Reiß and P. van Loock, “Deep reinforcement learning for key distribution based on quantum repeaters,”Phys. Rev. A, vol. 108, p. 012406, 2023

    2023

  13. [21]

    Deep reinforce- ment learning-based routing and resource assignment in quantum key distribution-secured optical networks,

    P. Sharma, S. Gupta, V . Bhatia, and S. Prakash, “Deep reinforce- ment learning-based routing and resource assignment in quantum key distribution-secured optical networks,”IET Quantum Commun., 2023

    2023

  14. [22]

    Adversarial learning game for intrusion detection in quantum key distribution,

    N. Mohamed and S. Al-Kuwari, “Adversarial learning game for intrusion detection in quantum key distribution,” arXiv:2603.03502, 2026

    2026

  15. [23]

    Adversarial limits of quantum certification: When Eve defeats detection,

    D. E. Tasar, “Adversarial limits of quantum certification: When Eve defeats detection,” arXiv:2512.04391, 2025

    2025

Pith tools

Reviewed June 26, 2026 · model on record in the stance chip above.