pith. sign in

arxiv: 2605.21002 · v1 · pith:BOLNEL4Tnew · submitted 2026-05-20 · 💻 cs.CR · cs.CV· cs.CY· cs.MM

Verifiable Provenance and Watermarking for Generative AI: An Evidentiary Framework for International Operational Law and Domestic Courts

Gustav Olaf Yunus Laitinen-Fredriksson Lundstr\"om-Imanov , Nurana Abdullayeva This is my paper

Pith reviewed 2026-05-21 04:13 UTC · model grok-4.3

classification 💻 cs.CR cs.CVcs.CYcs.MM
keywords generative AIcontent provenancewatermarkingevidentiary frameworkthreat modelzero-knowledge attestationlegal sufficiencyAI regulation
0
0 comments X

The pith

A unified evidentiary framework maps cryptographic provenance, statistical watermarking, and zero-knowledge attestation to proof requirements in international operational law, domestic courts, and AI product regulation.

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

Generative AI now creates photorealistic content at scale, undermining traditional methods for determining authenticity in legal and operational settings. The paper constructs a single framework that aligns technical mechanisms for tracing and verifying origins with the distinct evidentiary demands of three regimes: command decisions under the law of armed conflict, admissibility rules in criminal and civil procedure, and persistence audits under the EU Artificial Intelligence Act. It defines a five-tier threat model from simple regeneration to insider forgery, releases a public benchmark of 12000 generated items across modalities with 72000 evaluation samples under laundering pipelines, and evaluates representative schemes on detection performance. Empirical results on true positive rates, robustness curves, and overhead are then converted into regime-specific legal sufficiency scores. A sympathetic reader would care because the work supplies a shared reference that lets engineers, lawyers, and operators assess whether a given detection method meets the bar for real-world use without each community starting from scratch.

Core claim

The central claim is that cryptographic content provenance, robust statistical watermarking, and zero-knowledge attestation can be systematically mapped to the proof standards of international operational law, domestic procedure, and product regulation, with a public benchmark and five-tier threat model enabling empirical detection bounds to be translated into concrete legal sufficiency thresholds for command decisions, court admissibility, and regulatory compliance.

What carries the argument

The regime-conditioned legal sufficiency score that evaluates technical detection metrics against the requirements of each legal regime inside the five-tier threat model spanning naive regeneration through insider provenance forgery.

If this is right

  • Detection performance above a regime-specific threshold could justify reliance on AI-generated intelligence for targeting decisions under the law of armed conflict.
  • Courts could adopt the translated robustness bounds as one factor when deciding admissibility of synthetic media in civil or criminal cases.
  • Regulators could incorporate the benchmark results into audits that check whether generative AI products maintain verifiable provenance after distribution.
  • Developers could prioritize schemes that achieve high legal sufficiency scores rather than optimizing solely for technical detection rates.
  • Operators could use the reproducible pipeline to certify content chains that satisfy multiple overlapping legal regimes at once.

Where Pith is reading between the lines

These are editorial extensions of the paper, not claims the author makes directly.

  • The same mapping approach could be tested against additional regulatory frameworks outside the EU AI Act, such as emerging rules in other jurisdictions.
  • Real-world application might require hybrid processes that combine the automated scores with human review for the highest-stakes uses.
  • A practical next step would be to run selected schemes through simulated legal or operational reviews to check whether the translated thresholds hold under adversarial questioning.

Load-bearing premise

Quantitative detection metrics such as true positive rate at fixed false positive rate and robustness area under the curve can be translated directly into legal sufficiency thresholds without additional domain-specific validation in actual proceedings or operations.

What would settle it

A documented instance in which a method that scores above the framework's legal sufficiency threshold for a given regime is nevertheless rejected for evidentiary use in court or command review, or conversely a method below the threshold is accepted on other grounds.

Figures

Figures reproduced from arXiv: 2605.21002 by Gustav Olaf Yunus Laitinen-Fredriksson Lundstr\"om-Imanov, Nurana Abdullayeva.

Figure 1
Figure 1. Figure 1: Three legal regimes mapped to proof object components. [PITH_FULL_IMAGE:figures/full_fig_p003_1.png] view at source ↗
Figure 2
Figure 2. Figure 2: Proof object architecture. Four sources feed a regime conditioned Dempster Shafer aggregator that produces a single sufficiency score [PITH_FULL_IMAGE:figures/full_fig_p005_2.png] view at source ↗
Figure 3
Figure 3. Figure 3: Five tier adversary capability ladder used throughout the [PITH_FULL_IMAGE:figures/full_fig_p005_3.png] view at source ↗
Figure 4
Figure 4. Figure 4: ROC curves at adversary tier 2 (P1 and P2 laundering). [PITH_FULL_IMAGE:figures/full_fig_p006_4.png] view at source ↗
read the original abstract

Generative artificial intelligence now synthesizes photorealistic imagery, audio, and video at a cost that defeats traditional forensic intuition. The legal consequences span three regimes studied so far in isolation: international operational law, domestic procedure, and product regulation. This article presents a unified evidentiary framework that maps cryptographic content provenance, robust statistical watermarking, and zero knowledge attestation to the proof requirements of each regime. We define a five tier threat model spanning naive regeneration, adversarial laundering, cross model regeneration, active watermark removal, and insider provenance forgery. We release a public benchmark of 12000 generated items across image, audio, and video modalities under six laundering pipelines for 72000 evaluation samples. We evaluate four representative schemes and report true positive rate at fixed false positive rate, robustness area under the curve, computational overhead, and a regime conditioned legal sufficiency score. We translate empirical detection bounds into legal sufficiency thresholds for command decisions under the law of armed conflict, for criminal and civil admissibility under domestic procedure, and for persistence audits under the European Union Artificial Intelligence Act and analogous regimes. The result is a reproducible reference pipeline, a public benchmark, and model annexes that lawyers, engineers, and operators can deploy together.

Editorial analysis

A structured set of objections, weighed in public.

Desk editor's note, referee report, simulated authors' rebuttal, and a circularity audit. Tearing a paper down is the easy half of reading it; the pith above is the substance, this is the friction.

Referee Report

2 major / 2 minor

Summary. The manuscript proposes a unified evidentiary framework that maps cryptographic content provenance, robust statistical watermarking, and zero-knowledge attestation to the proof requirements of international operational law (LOAC), domestic criminal/civil procedure, and product regulation regimes such as the EU AI Act. It introduces a five-tier threat model, releases a public benchmark consisting of 12,000 generated items across image/audio/video modalities evaluated under six laundering pipelines (yielding 72,000 samples), evaluates four representative schemes, and reports TPR at fixed FPR, robustness AUC, computational overhead, plus a regime-conditioned legal sufficiency score that translates empirical detection bounds into legal thresholds for command decisions, admissibility, and audits.

Significance. If the empirical-to-legal mappings can be substantiated, the work would offer a practical bridge between technical detection capabilities and legal proof standards, with the public benchmark, reproducible pipeline, and model annexes serving as concrete resources for lawyers, engineers, and operators. The large-scale evaluation across modalities and laundering attacks strengthens the technical foundation and enables falsifiable follow-on testing.

major comments (2)
  1. [Section 5] Section 5 (Legal Sufficiency Translation): The regime-conditioned legal sufficiency score is constructed by directly mapping TPR at fixed FPR and robustness AUC values to sufficiency thresholds for LOAC command decisions, domestic admissibility, and EU AI Act persistence audits. No derivation from cited precedents, Daubert-style reliability criteria, or simulated proceedings is provided, making the thresholds an authorial overlay rather than a validated evidentiary bridge; this mapping is load-bearing for the central claim of a unified framework.
  2. [§4.2 and Table 3] §4.2 and Table 3: The evaluation reports legal sufficiency scores for the four schemes, yet the manuscript does not include statistical significance testing, confidence intervals, or sensitivity analysis on the 72,000 samples when converting AUC/TPR into the legal thresholds; without these, the cross-regime comparisons rest on point estimates whose stability is unverified.
minor comments (2)
  1. [Abstract and §6] The abstract and §6 refer to 'model annexes' and a 'public benchmark' but the manuscript lacks an explicit persistent identifier, repository link, or instructions for accessing the 12,000-item dataset and evaluation code, which would improve reproducibility.
  2. [§3] Notation for the five-tier threat model (naive regeneration through insider forgery) is introduced in §3 but not consistently cross-referenced when discussing which tiers are covered by the six laundering pipelines in the benchmark; a summary table would clarify coverage.

Simulated Author's Rebuttal

2 responses · 0 unresolved

We thank the referee for the constructive feedback and for recognizing the potential of the unified evidentiary framework. We address each major comment below with specific revisions to strengthen the legal mappings and empirical validations.

read point-by-point responses

  1. Referee: [Section 5] Section 5 (Legal Sufficiency Translation): The regime-conditioned legal sufficiency score is constructed by directly mapping TPR at fixed FPR and robustness AUC values to sufficiency thresholds for LOAC command decisions, domestic admissibility, and EU AI Act persistence audits. No derivation from cited precedents, Daubert-style reliability criteria, or simulated proceedings is provided, making the thresholds an authorial overlay rather than a validated evidentiary bridge; this mapping is load-bearing for the central claim of a unified framework.

    Authors: We agree that Section 5 would benefit from an explicit derivation of the sufficiency thresholds. The current mappings draw from the proof burdens described in the cited sources (Additional Protocol I for LOAC command responsibility, Daubert v. Merrell Dow for reliability factors in domestic admissibility, and EU AI Act Articles 50-52 for audit persistence), but the manuscript does not walk through the translation step by step. In revision we will add a new subsection 5.1 that derives each threshold from these precedents, applies Daubert-style criteria to the reported TPR/FPR and AUC values, and includes a brief simulated proceeding example for each regime. This will make the bridge evidentiary rather than conceptual. revision: yes

  2. Referee: [§4.2 and Table 3] §4.2 and Table 3: The evaluation reports legal sufficiency scores for the four schemes, yet the manuscript does not include statistical significance testing, confidence intervals, or sensitivity analysis on the 72,000 samples when converting AUC/TPR into the legal thresholds; without these, the cross-regime comparisons rest on point estimates whose stability is unverified.

    Authors: The observation is accurate: the current version reports point estimates for the legal sufficiency scores without accompanying statistical tests or intervals. We will revise §4.2 to include bootstrap-derived 95% confidence intervals on the AUC and TPR values across the 72,000 samples, together with a sensitivity analysis that perturbs laundering pipeline parameters (e.g., compression quality, adversarial strength). Updated Table 3 will display these intervals alongside the scores so that cross-regime comparisons rest on quantified stability rather than point estimates alone. revision: yes

Circularity Check

0 steps flagged

No significant circularity; framework applies external methods to new domain

full rationale

The paper defines a threat model, releases a benchmark of 12000 items evaluated across 72000 samples, reports standard empirical metrics (TPR at fixed FPR, robustness AUC, overhead) for four schemes, and proposes translations of those metrics into regime-conditioned legal sufficiency thresholds. No equation or step reduces by construction to a fitted parameter or self-defined quantity; the legal mapping is presented as an interpretive bridge rather than a tautology. No load-bearing self-citation chain or uniqueness theorem imported from the authors' prior work is evident in the provided text. The derivation remains self-contained against external cryptographic and statistical benchmarks.

Axiom & Free-Parameter Ledger

0 free parameters · 1 axioms · 0 invented entities

The framework rests on standard cryptographic and statistical assumptions plus one domain assumption that legal evidentiary standards can be operationalized via detection performance numbers.

axioms (1)
  • domain assumption Legal proof requirements across the studied regimes can be quantified using technical detection metrics such as TPR at fixed FPR and robustness AUC.
    This premise is required to translate empirical bounds into legal sufficiency thresholds as described in the abstract.

pith-pipeline@v0.9.0 · 5776 in / 1335 out tokens · 63812 ms · 2026-05-21T04:13:47.210563+00:00 · methodology

discussion (0)

Sign in with ORCID, Apple, or X to comment. Anyone can read and Pith papers without signing in.

Lean theorems connected to this paper

Citations machine-checked in the Pith Canon. Every link opens the source theorem in the public Lean library.

What do these tags mean?
matches
The paper's claim is directly supported by a theorem in the formal canon.
supports
The theorem supports part of the paper's argument, but the paper may add assumptions or extra steps.
extends
The paper goes beyond the formal theorem; the theorem is a base layer rather than the whole result.
uses
The paper appears to rely on the theorem as machinery.
contradicts
The paper's claim conflicts with a theorem or certificate in the canon.
unclear
Pith found a possible connection, but the passage is too broad, indirect, or ambiguous to say the theorem truly supports the claim.

Reference graph

Works this paper leans on

49 extracted references · 49 canonical work pages

  1. [1]

    Deepfake detection by human crowds, machines, and machine-informed crowds,

    M. Groh, Z. Epstein, C. Firestone, and R. Picard, “Deepfake detection by human crowds, machines, and machine-informed crowds,”Proceedings of the National Academy of Sciences, vol. 119, no. 1, p. e2110013119, 2022

    work page 2022

  2. [2]

    AI-synthesized faces are indistin- guishable from real faces and more trustworthy,

    S. J. Nightingale and H. Farid, “AI-synthesized faces are indistin- guishable from real faces and more trustworthy,”Proceedings of the National Academy of Sciences, vol. 119, no. 8, p. e2120481119, 2022

    work page 2022

  3. [3]

    [Online]

    Coalition for Content Provenance and Authenticity,C2PA Tech- nical Specification, Version 2.1 or later, 2024. [Online]. Available: https://c2pa.org/specifications

    work page 2024

  4. [4]

    CNN-generated images are surprisingly easy to spot for now,

    S. Y. Wang, O. Wang, R. Zhang, A. Owens, and A. A. Efros, “CNN-generated images are surprisingly easy to spot for now,” inProc. IEEE/CVF Conf. Computer Vision and Pattern Recog- nition, 2020, pp. 8695 to 8704

    work page 2020

  5. [5]

    Towards universal fake image detectors that generalize across generative models,

    U. Ojha, Y. Li, and Y. J. Lee, “Towards universal fake image detectors that generalize across generative models,” inProc. IEEE/CVF Conf. Computer Vision and Pattern Recognition, 2023, pp. 24480 to 24489

    work page 2023

  6. [6]

    The stable signature: Rooting watermarks in latent diffusion models,

    P. Fernandez, G. Couairon, H. Jégou, M. Douze, and T. Furon, “The stable signature: Rooting watermarks in latent diffusion models,” inProc. IEEE/CVF Int. Conf. Computer Vision, 2023, pp. 22466 to 22477

    work page 2023

  7. [7]

    Tree- Ring watermarks: Fingerprints for diffusion images that are invis- ible and robust,

    Y. Wen, J. Kirchenbauer, J. Geiping, and T. Goldstein, “Tree- Ring watermarks: Fingerprints for diffusion images that are invis- ible and robust,” inAdvances in Neural Information Processing Systems, vol. 36, 2023

    work page 2023

  8. [8]

    Gaussian Shading: Provable performance-lossless image watermarking for diffusion models,

    Z. Yang, K. Zeng, K. Chen, H. Fang, W. Zhang, and N. Yu, “Gaussian Shading: Provable performance-lossless image watermarking for diffusion models,” inProc. IEEE/CVF Conf. Computer Vision and Pattern Recognition, 2024

    work page 2024

  9. [9]

    A watermark for large language models,

    J. Kirchenbauer, J. Geiping, Y. Wen, J. Katz, I. Miers, and T. Goldstein, “A watermark for large language models,” inProc. Int. Conf. Machine Learning, 2023

    work page 2023

  10. [10]

    Robust distortion-free watermarks for language models,

    R. Kuditipudi, J. Thickstun, T. Hashimoto, and P. Liang, “Robust distortion-free watermarks for language models,” arXiv:2307.15593, 2023

    work page arXiv 2023

  11. [11]

    Invisible image watermarks are provably remov- able using generative AI,

    X. Zhao et al., “Invisible image watermarks are provably remov- able using generative AI,” inAdvances in Neural Information Processing Systems, vol. 37, 2024. IEEE TRANSACTIONS ON INFORMATION FORENSICS AND SECURITY, VOL. XX, NO. X, 2026 13

    work page 2024

  12. [12]

    Robustness of AI-image detectors: Funda- mental limits and practical attacks,

    M. Saberi et al., “Robustness of AI-image detectors: Funda- mental limits and practical attacks,” inInt. Conf. Learning Representations, 2024

    work page 2024

  13. [13]

    WAVES: Benchmarking the robustness of image watermarks,

    B. An et al., “WAVES: Benchmarking the robustness of image watermarks,” inProc. Int. Conf. Machine Learning, 2024

    work page 2024

  14. [14]

    Scaling up trustless DNN inference with zero-knowledge proofs,

    D. Kang, T. Hashimoto, I. Stoica, and Y. Sun, “Scaling up trustless DNN inference with zero-knowledge proofs,” arXiv:2210.08674, 2022

    work page arXiv 2022

  15. [15]

    zkLLM: Zero knowledge proofs for large language models,

    H. Sun, J. Li, and H. Zhang, “zkLLM: Zero knowledge proofs for large language models,” inProc. ACM Conf. Computer and Communications Security, 2024

    work page 2024

  16. [16]

    High-speed high-security signatures,

    D. J. Bernstein, N. Duif, T. Lange, P. Schwabe, and B. Y. Yang, “High-speed high-security signatures,”Journal of Cryptographic Engineering, vol. 2, no. 2, pp. 77 to 89, 2012

    work page 2012

  17. [17]

    Shafer,A Mathematical Theory of Evidence

    G. Shafer,A Mathematical Theory of Evidence. Princeton, NJ: Princeton University Press, 1976

    work page 1976

  18. [18]

    LAION-5B: An open large-scale dataset for training next generation image-text models,

    C. Schuhmann et al., “LAION-5B: An open large-scale dataset for training next generation image-text models,” inAdvances in Neural Information Processing Systems, vol. 35, 2022

    work page 2022

  19. [19]

    CommonVoice:Amassively-multilingualspeech corpus,

    R.Ardilaetal.,“CommonVoice:Amassively-multilingualspeech corpus,”inProc.LanguageResourcesandEvaluationConference, 2020

    work page 2020

  20. [20]

    Protocol Additional to the Geneva Conventions of 12 August 1949, and relating to the Protection of Victims of International Armed Conflicts (Protocol I), 8 June 1977, 1125 UNTS 3

    work page 1949

  21. [21]

    Doswald-Beck, Ed.,San Remo Manual on International Law Applicable to Armed Conflicts at Sea

    L. Doswald-Beck, Ed.,San Remo Manual on International Law Applicable to Armed Conflicts at Sea. Cambridge: Cambridge University Press, 1995

    work page 1995

  22. [22]

    M. N. Schmitt, Ed.,Tallinn Manual 2.0 on the International Law Applicable to Cyber Operations, 2nd ed. Cambridge: Cambridge University Press, 2017

    work page 2017

  23. [23]

    Targeting and international humanitarian law in Afghanistan,

    M. N. Schmitt, “Targeting and international humanitarian law in Afghanistan,”International Law Studies, vol. 85, pp. 307 to 339, 2009

    work page 2009

  24. [24]

    Henderson,The Contemporary Law of Targeting

    I. Henderson,The Contemporary Law of Targeting. Leiden: Martinus Nijhoff, 2009

    work page 2009

  25. [25]

    W. H. Boothby,The Law of Targeting. Oxford: Oxford University Press, 2014

    work page 2014

  26. [26]

    Geiss and H

    R. Geiss and H. Lahmann, Eds.,Research Handbook on Warfare and Artificial Intelligence. Cheltenham: Edward Elgar, 2021

    work page 2021

  27. [27]

    Deep fakes: A looming challenge for privacy, democracy, and national security,

    R. Chesney and D. K. Citron, “Deep fakes: A looming challenge for privacy, democracy, and national security,”California Law Review, vol. 107, pp. 1753 to 1819, 2019

    work page 2019

  28. [28]

    Rewriting the right of publicity for synthetic media,

    M. Pavis, “Rewriting the right of publicity for synthetic media,” Stanford Technology Law Review, vol. 25, 2021

    work page 2021

  29. [29]

    Federal Rules of Evidence, Rule 901 (Authenticating or Identify- ing Evidence), United States, 2024 edition

    work page 2024

  30. [30]

    Electronic evidence and the federal rules,

    D. J. Capra and L. R. Richter, “Electronic evidence and the federal rules,”Fordham Law Review, vol. 86, pp. 1597 to 1632, 2017

    work page 2017

  31. [31]

    Regulation (EU) 2024/1183 of the European Parliament and of the Council of 11 April 2024 amending Regulation (EU) No 910/2014 (eIDAS 2),Official Journal of the European Union, L, 30 April 2024

    work page 2024

  32. [32]

    Regulation (EU) 2024/1689 of the European Parliament and of the Council of 13 June 2024 laying down harmonised rules on artificial intelligence (Artificial Intelligence Act),Official Journal of the European Union, L, 12 July 2024

    work page 2024

  33. [33]

    National Institute of Standards and Technology,Reducing Risks Posed by Synthetic Content: An Overview of Technical Approaches to Digital Content Transparency, NIST AI 100-4, 2024

    work page 2024

  34. [34]

    Mason and D

    S. Mason and D. Seng, Eds.,Electronic Evidence, 4th ed. London: University of London Press, 2017

    work page 2017

  35. [35]

    C. S. D. Brown,Digital Forensic Evidence in the Courtroom: Understanding Content and Quality, 3rd ed. Routledge, 2020

    work page 2020

  36. [36]

    Regulating ChatGPT and other large generative AI models,

    P. Hacker, A. Engel, and M. Mauer, “Regulating ChatGPT and other large generative AI models,” inProc. ACM Conf. Fairness, Accountability, and Transparency, 2023

    work page 2023

  37. [37]

    EU Artificial Intelligence Act: The European approach to AI,

    M. Kop, “EU Artificial Intelligence Act: The European approach to AI,” Stanford-Vienna Transatlantic Technology Law Forum, Stanford Law School, 2021

    work page 2021

  38. [38]

    Demystifying the Draft EU Artificial Intelligence Act,

    M. Veale and F. Z. Borgesius, “Demystifying the Draft EU Artificial Intelligence Act,”Computer Law Review International, vol. 22, no. 4, pp. 97 to 112, 2021

    work page 2021

  39. [39]

    Roscini,Cyber Operations and the Use of Force in Interna- tional Law

    M. Roscini,Cyber Operations and the Use of Force in Interna- tional Law. Oxford: Oxford University Press, 2014

    work page 2014

  40. [40]

    P. O. Ekelöf, R. Boman, and H. Edelstam,Rättegång. Fjärde häftet, 7th ed. Stockholm: Norstedts Juridik, 2009

    work page 2009

  41. [41]

    Mahmoud Mustafa Busayf Al-Werfalli, ICC-01/11-01/17, 15 August 2017

    International Criminal Court, Pre-Trial Chamber I,Warrant of Arrest, Prosecutor v. Mahmoud Mustafa Busayf Al-Werfalli, ICC-01/11-01/17, 15 August 2017

    work page 2017

  42. [42]

    United Nations Office of the High Commissioner for Human Rights and Human Rights Center, UC Berkeley School of Law, Berkeley Protocol on Digital Open Source Investigations, United Nations Publication, 2022

    work page 2022

  43. [43]

    United States Federal Trade Commission,In the Matter of Rytr LLC, Decision and Order, Operation AI Comply, 2024

    work page 2024

  44. [44]

    Regulation (EU) 2022/2065 of the European Parliament and of the Council of 19 October 2022 on a Single Market For Digital Services (Digital Services Act),Official Journal of the European Union, L 277, 27 October 2022

    work page 2022

  45. [45]

    The White House, Executive Order 14110 on the Safe, Secure, and Trustworthy Development and Use of Artificial Intelligence, 30 October 2023

    work page 2023

  46. [46]

    International Organization for Standardization,ISO/IEC DIS 22144 Digital media provenance, draft international standard, 2025

    work page 2025

  47. [47]

    Code of Criminal Procedure of the Republic of Azerbaijan, Articles 124 and 128 (electronic evidence), as amended through 2024

    work page 2024

  48. [48]

    Forensic science reform in the 21st century: A major conference, a blockbuster report, and reasons to be pessimistic,

    J. J. Koehler, “Forensic science reform in the 21st century: A major conference, a blockbuster report, and reasons to be pessimistic,”Law, Probability and Risk, vol. 13, no. 1, pp. 1 to 5, 2014

    work page 2014

  49. [49]

    Deepfake of Zelensky surrender shared on hacked Ukraine news,

    J. Vincent, “Deepfake of Zelensky surrender shared on hacked Ukraine news,”The Verge, 16 March 2022. Gustav Olaf Y unus Laitinen-F redriksson Lundström-Imanov (Student Member, IEEE) is an LL.M. candidate in International Operational Law at Försvarshögskolan (Swedish Defence University). His research interests include the law of armed conflict, information...

    work page 2022