Draft:ZKAP

  • Comment: Submitted for review. Article concerns a legal-engineering methodology with two Bulgarian patent applications, four Zenodo deposits with DOIs, an ISSN-indexed conference paper, an institutional publication on the European Commission community platform, and direct comparison with three independently published academic frameworks (ZKMLOps, ZKAUDIT, IoA Audit). The author of the article (User:Sirrado) is the inventor of the method and has declared a conflict of interest on the user page. Sirrado (talk) 08:17, 5 May 2026 (UTC)
For the cryptographic-token authentication scheme in blockchain and storage systems, see Zero-Knowledge Access Pass.
Radoslav Y. Radoslavov, inventor of ZKAP (2026)

ZKAP (Zero-Knowledge Audit Protocol) is a legal-engineering methodology for cryptographic verification of regulatory compliance of artificial intelligence systems, introduced by Bulgarian jurist Radoslav Yordanov Radoslavov in the period 2025–2026, and the subject of two Bulgarian patent applications.[1][2] The protocol applies zero-knowledge proofs to allow a regulator to obtain mathematical assurance that a given system satisfies a formally specified regulatory requirement, without disclosing the sensitive data, model parameters or internal state on which the assurance rests.

ZKAP in this sense is distinct from other cryptographic schemes that share the same acronym, in particular Zero-Knowledge Access Pass — an anonymous-authorisation token used in distributed-storage and blockchain systems. Both approaches are built on the mathematics of zero-knowledge proofs, but they address different problems: Zero-Knowledge Access Pass establishes a right of access without revealing the user's identity, whereas the Zero-Knowledge Audit Protocol establishes regulatory compliance without revealing the regulated data.

History

The concept was first presented in March 2026 at the XI International Scientific Conference "High Technologies. Business. Society" in Borovets, Bulgaria.[1] A preprint with full cryptographic formalisation and a soundness theorem was deposited on Zenodo on 22 April 2026.[3] Two further pieces appeared in the European Commission's Apply AI Alliance community-content platform (Futurium) on 2 April and 18 April 2026, presenting the method to a regulatory audience.[4][5]

The conceptual core of the method is described in an earlier 2025 publication by Radoslavov in Industry 4.0 (Winter Session, vol. 2), where the dual-domain architecture and the idea of cryptographic verification of legal duties are introduced, but without specification of the full protocol.[2]

Regulatory problem

ZKAP is a methodological response to a structural conflict between three regulatory regimes of the European Union:

  • Regulation (EU) 2024/1689 (the AI Act) requires transparency of high-risk AI systems — explainability of decisions, traceability of training data, risk assessment, active human oversight, and immutable audit logs;
  • The General Data Protection Regulation (GDPR; Regulation (EU) 2016/679) prohibits disclosure of personal data without a valid legal basis;
  • Directive (EU) 2016/943 protects trade secrets and undisclosed know-how of model developers.[4]

For cognitively opaque models — in particular large language models — the AI Act's transparency duties and the GDPR / trade-secret prohibitions enter into direct collision: full disclosure as required by the supervisor simultaneously violates the other two frameworks. The same conflict structure arises under the NIS2 Directive in respect of cybersecurity duties.[5]

Core idea

The protocol substitutes a duty of disclosure with a duty of cryptographic proof: the regulator receives a mathematical assurance that the system is compliant with a particular rule, without seeing the data, the model weights, or the internal state on which the rule operates.[2][1] The proof is a zero-knowledge proof — a cryptographic construction known since the 1980s, which allows one party to prove the truth of a statement without revealing the underlying content.[1]

Architecture: R-Domain and C-Domain

The methodology partitions information in the system into two zones:

  • R-Domain (Raw Domain) contains raw data — training sets, model weights, end-user personal data, internal policies. This zone never leaves the obligated party's infrastructure.
  • C-Domain (Compliance Domain) contains the cryptographic proofs derived from the R-Domain. This zone is publicly visible to the regulator. The proofs do not disclose the contents of the R-Domain but mathematically attest to specific statements about it.[1]

According to the publications, sensitive information consequently does not travel along the conventional chain "regulator → court → experts → opposing party", because such information does not exist outside the obligated party's perimeter.[2]

Polynomialisation of law

The conceptual core of ZKAP is the so-called "polynomialisation of law" — the translation of formalised legal norms into mathematical constraints, which can be verified by means of zero-knowledge proofs (such as zk-SNARKs or zk-STARKs).[1] An obligation expressed in legal language is transposed into an obligation expressed in the language of cryptography, so that verification becomes machine-executable and independent of human discretion.

Hardware anchoring

The publications describe the possibility of embedding basic formalised regulatory rules directly into hardware — at the level of the computational element, rather than as a subsequent software control layer. The implementation uses a specialised hardware element designated Provable Arithmetic Logic Unit (pALU), which secures determinism and resistance to algorithmic drift. This approach realises the principle of "safe by design": violation of a regulatory invariant cannot occur during execution, instead of being detected after the fact.[1]

Operating modes

The protocol is presented as operating in several modes depending on the system's risk profile and regulatory context:[1][5]

  • Embedded mode (compliance-by-design) — a compliance proof is generated on every execution of the system, in real time.
  • Retrospective mode — subsequent verification of past operations against retained logs.
  • Parallel mode (shadow audit) — independent oversight in real time, without intrusion into the production environment.
  • Hybrid mode — a staged combination, suited to administrative and judicial procedures.

Inventor

ZKAP is the work of Bulgarian jurist Radoslav Yordanov Radoslavov (ORCID: 0009-0003-6868-8083). The methodology was developed independently in 2025–2026 on the basis of a dual practical engagement — in civil forfeiture practice and in legal regulation of artificial intelligence. Both fields share a common structural problem: a duty to disclose imposed by a supervising authority versus a prohibition against disclosure of sensitive material.[2][1]

Between 2006 and 2023, Radoslavov served as a jurist in the Bulgarian anti-corruption commission (successively known as CECDPMI, KONPI, and the Commission for Counteracting Corruption and Forfeiture of Illegally Acquired Property) — the institutional context on which the conceptual foundations of ZKAP were later built.[6]

Patents

The principal technical constructions of ZKAP are the subject of two Bulgarian patent applications:[1]

  • BG/P/2026/114317 — Bulgarian Patent Office, filed 30 March 2026. International Patent Classification (IPC): G06F 21/64; G06N 20/00; H04L 9/32.
  • BG/P/2026/114328 (system identifier PTBG202600000316742) — Bulgarian Patent Office, filed 12 April 2026; partial priority from BG/P/2026/114317.

Both filings activate Paris Convention priority until 30 March 2027 and 12 April 2027, respectively, for international extension via the European Patent Office, the UK Intellectual Property Office, and the Patent Cooperation Treaty.

Publications

Peer-reviewed papers

  • Radoslavov, R. Y. (2025). "Management and Regulation of Artificial Intelligence Models: Concept for Transparency and Accountability in Administrative Activities". Industry 4.0, Winter Session, vol. 2, pp. 320–321. ISSN 2534-997X. DOI: 10.5281/zenodo.19614243.
  • Radoslavov, R. Y. (2026). "Management and Regulation of AI Models in Public Administration: Cryptographic Transparency and Digitalization of Legal Norms". Artificial Intelligence Proceedings, vol. 2026, no. 1, pp. 75–78. ISSN 3033-2923 / 3134-1667. DOI: 10.5281/zenodo.19509511.

Preprint

  • Radoslavov, R. Y. (2026). "ZKAP: An Enforcement Protocol for Verifiable Regulatory Compliance of Machine-Learning Inference via Certified Stack Binding". Preprint, Zenodo, 22 April 2026 (under embargo until 31 March 2027). DOI: 10.5281/zenodo.19698949.

Authored contributions on the European Commission community platform

ZKAP forms part of a broader academic line which applies zero-knowledge cryptography to the auditing of machine learning. Related frameworks include:

  • ZKMLOps (Scaramuzza et al., 2025) — an engineering framework for integrating zero-knowledge primitives into MLOps pipelines.[7]
  • ZKAUDIT (Waiwitlikhit et al., 2024) — a computation-focused framework for zero-knowledge audit of individual ML inference computations.
  • Zero-Knowledge Audit for Internet of Agents (Jing & Qi, 2025) — application of zero-knowledge proofs to inter-agent communication.[8]

ZKAP is distinguished from these by its focus on the legal-engineering layer — specific articles of the AI Act (Articles 12, 13, 14, 15, 27) and the architectural separation between R-Domain and C-Domain — rather than on individual engineering components.[1]

References

  1. ^ a b c d e f g h i j k Radoslavov, R. Y. (2026). "Management and Regulation of Artificial Intelligence Models in Public Administration: Cryptographic Transparency and Digitalization of Legal Norms". Artificial Intelligence Proceedings, vol. 2026, no. 1, pp. 75–78. ISSN 3033-2923 / 3134-1667. XI International Scientific Conference "High Technologies. Business. Society", Borovets, Bulgaria, 23–26 March 2026. DOI: 10.5281/zenodo.19509511.
  2. ^ a b c d e Radoslavov, R. Y. (2025). "Management and Regulation of Artificial Intelligence Models: Concept for Transparency and Accountability in Administrative Activities". Industry 4.0, Winter Session, vol. 2, pp. 320–321. ISSN 2534-997X. DOI: 10.5281/zenodo.19614243.
  3. ^ Radoslavov, R. Y. (2026). "ZKAP: An Enforcement Protocol for Verifiable Regulatory Compliance of Machine-Learning Inference via Certified Stack Binding". Preprint, Zenodo, 22 April 2026 (under embargo until 31 March 2027, coinciding with the Paris Convention priority deadline of the underlying Bulgarian patent application). DOI: 10.5281/zenodo.19698949.
  4. ^ a b European Commission, Apply AI Alliance (Futurium). "ZKAP: Zero-Knowledge Audit Protocol — Solving the Cognitive Barrier in AI Act & NIS2 Oversight", 2 April 2026.
  5. ^ a b c European Commission, Apply AI Alliance (Futurium). "After Mythos: Why Frontier AI Conformity Assessment Requires a Cryptographic Layer", 18 April 2026.
  6. ^ Bulgarian Anti-Corruption Commission Registry, 2006–2023. https://anticorruption.government.bg/.
  7. ^ Scaramuzza, M.; Cordeiro Ferreira, R.; Suller, R.; Quattrocchi, G.; Tamburri, D. A.; van den Heuvel, W.-J. (2025). "Show Me You Comply… Without Showing Me Anything: Zero-Knowledge Software Auditing for AI-Enabled Systems". arXiv preprint arXiv:2510.26576. https://arxiv.org/abs/2510.26576.
  8. ^ Jing, Y.; Qi, X. (2025). "Zero-Knowledge Audit for Internet of Agents: Privacy-Preserving Communication Verification with Model Context Protocol". arXiv preprint arXiv:2512.14737. https://arxiv.org/abs/2512.14737.

Category:Artificial intelligence Category:Cryptography Category:Zero-knowledge proofs Category:Technology law Category:Artificial intelligence regulation Category:Bulgarian inventions

Content Disclaimer

Informasi ini disarikan dari Wikipedia dan disajikan kembali untuk tujuan edukasi. Konten tersedia di bawah lisensi CC BY-SA 3.0. Kami tidak bertanggung jawab atas ketidakakuratan data yang bersumber dari kontribusi publik tersebut.

  1. The information displayed on this website is sourced in part or in whole from Wikipedia and has been adapted for the purpose of restating it. We strive to provide accurate and relevant information, however:
  2. There is no guarantee of absolute accuracy. Wikipedia is an open, collaborative project that can be edited by anyone, so information is subject to change.
  3. It is not intended to constitute professional advice. The content displayed is for informational and educational purposes only. For important decisions (e.g., medical, legal, or financial), please consult a professional.
  4. Content copyright. Wikipedia is licensed under the Creative Commons Attribution-ShareAlike License (CC BY-SA). This means that content may be reused with appropriate attribution and shared under a similar license.
  5. Responsible use. Any risk arising from the use of information from this website is entirely the responsibility of the user.