The DOI matches its Crossref record; 21 of the 22 references it cites were found in a registry or an index; 11 of the 54 sources its claims cite were checked for support.
AI Supply Chain Security: MBOM-PQC Provenance, PQC Attestation, and a Maturity Model for Quantum-Resistant Assurance
Open accessAbstract
A formal Model Bill of Materials with PQC-safe extensions, a unified signing and attestation pipeline, and a maturity model are proposed for AI supply chain integrity.
Overview
CoChat Summary
TL;DR
The paper synthesizes 54 policy, standards, academic, and incident sources to argue that AI supply chains lack verifiable provenance and that classical signatures used for model lineage and attestation will fail against "harvest-now, forge-later" quantum attacks. It proposes three prescriptive artifacts: the MBOM-PQC provenance schema, a PQC-safe signing and attestation pipeline built on ML-DSA/SLH-DSA and hybrid signatures, and the five-level SCAMM maturity model.
Key findings
- Existing frameworks (NIST AI RMF, NIST SSDF, SP 800-204D, DoD CDAO RAI Toolkit, SPDX/CycloneDX SBOM standards, and PQC transition guidance such as SP 800-208 and CNSA 2.0) address adjacent concerns but none defines a verifiable AI model provenance structure or PQC-safe signing/attestation pipeline.
- The evidence synthesis screened 142 records, deduplicated to 123, screened to 76 full-text reviews, and included 54 sources: 31 policy/standards documents, 15 peer-reviewed academic papers, and 8 documented supply chain incidents (28 of the 54 are cited directly in the manuscript body).
- The MBOM-PQC schema is organized into seven components (model metadata; pre-training dataset lineage; pre-trained model dependencies; fine-tuning artifacts; training environment and pipeline; deployment packaging; cryptographic integrity fields), with three PQC extensions: hybrid signature bundles, PQC-safe certificate chains, and long-term integrity anchors.
- The signing pipeline defines five stages (Ingestion, Verification, Signing, Attestation, Deployment) and recommends ML-DSA-65 as the default parameter set, citing a signature size of 3,309 bytes (~3.3 KB) versus 64 bytes for Ed25519, with SLH-DSA signatures ranging roughly 8–50 KB depending on parameter set; SCAMM defines five cumulative levels from Ad Hoc (L1) to Continuously Attested/Zero Trust-aligned (L5).
Methods
- Dual-method design: systematic evidence synthesis (PRISMA 2020 adapted for policy, standards, and security-incident sources) combined with prescriptive architectural development; the framework artifacts are design-science outputs intended for future operational validation rather than empirical evaluation.
- Evidence collected January–December 2025 across five source classes (standards bodies, government policy, incident records, academic literature, industry reports) via searches on IEEE Xplore, ACM Digital Library, IACR ePrint, NIST CSRC, NSA.gov, CISA.gov, and curated incident repositories; inclusion required 2020–2026 publication, relevance to AI supply chain risk/cryptographic integrity/PQC transition, normative or technical content, and extractable provenance/signing/attestation content, with four exclusion codes (E1–E4).
- Sources were coded with a nine-field extraction template and assigned confidence tiers (A/B/C); extracted requirements were mapped to schema fields, pipeline components, and SCAMM indicators via traceability matrices (Tables 1 and 2), with full materials in Supplementary Materials.
Limitations
- Stated: PQC algorithms and certificate formats continue to evolve, so specific parameter-set recommendations may need revision as NIST/IETF guidance matures; the framework lacks empirical validation through large-scale operational deployments; it depends on upstream providers supplying sufficient metadata and signatures, which varies widely; and continuous learning systems create high-velocity signing/attestation overhead that may require batched, delta-signing, or checkpoint-based adaptations.
- Evident from the text: the version provided is a Preprints.org preprint explicitly marked "Article Not peer-reviewed version," so it has not undergone peer review; the SCAMM dashboard (Figure 10) is described as illustrative/representative rather than derived from a real organizational assessment; and the SHA-256 hash shown in Figure 6 is noted as illustrative only.
Model-written summary. Check it against the paper before relying on it.
Integrity signals
Registry
Cites
Statistics
Verification
Claims the paper makes
2 Valid · 21 Not valid · 31 Not checked · 81 Own results and interpretations
None of the sources cited for this claim have been checked yet.
In the paper
“PQC transition guidance, including NIST SP 800-208 [15] and CNSA 2.0 [8], does not address AI artifacts or model provenance.”
What the sources say
- Could not verifycites [15] Jiang, W., Synovic, N., Hyatt, M., Schorlemmer, T.R., Sethi, R., Lu, Y.-H., Thiruvathukal, G.K., and Davis, J.C. (2023). An Empirical Study of Pre-Trained Model Reuse in the Hugging Face Deep Learning Model Registry. Proceedings of the 2023 IEEE/ACM 45th International Conference on Software Engineering (ICSE), IEEE. sourceExperimental — accuracy not yet measured
The Abstract discusses provenance as an attribute of pre-trained models, but does not discuss whether PQC transition guidance addresses AI artifacts or model provenance. Searches for NIST SP 800-208, CNSA 2.0, and post-quantum guidance found no passage addressing that claim. The printed sentence cites [15] and [8], not this paper. (LLM, conf=0.73)
- Not checked yetcites [8] OWASP (2026, January 15). Top 10 for Large Language Model Applications, Version 2025: LLM03—Supply Chain; OWASP GenAI Security Project. Available online: https://genai.owasp.org/.
plain- Asserted: passes — The authors put it forward as true.
- Checkable: passes — It can be checked against evidence.
- Load-bearing: passes — The paper’s conclusions depend on it.
- Based: passes — The paper gives a reason why it holds.
- Could not verify
A statement the paper makes itself — it cites no source for it, so there is none to check. The quote shows it is in the paper.
In the paper
“Classical digital signatures used to verify model lineage, dataset integrity, and pipeline attestation will not provide long-term protection against quantum-enabled forgery.”
plain- Asserted: passes — The authors put it forward as true.
- Checkable: passes — It can be checked against evidence.
- Load-bearing: passes — The paper’s conclusions depend on it.
- Based: passes — The paper gives a reason why it holds.
- Rests on an assumption not shown here.
In the paper
“Each component introduces potential attack vectors [10].”
What the sources say
- Not checked yetcites [10] NIST (2022). Secure Software Development Framework (SSDF), Version 1.1; SP 800-218.
plainstates a possible cause- Asserted: passes — The authors put it forward as true.
- Checkable: passes — It can be checked against evidence.
- Load-bearing: passes — The paper’s conclusions depend on it.
- Based: passes — The paper gives a reason why it holds.
A statement the paper makes itself — it cites no source for it, so there is none to check. The quote shows it is in the paper.
In the paper
“Training-time attacks-including data poisoning, model poisoning, and backdoor insertion-can compromise model behavior before deployment.”
plainstates a cause- Asserted: passes — The authors put it forward as true.
- Checkable: passes — It can be checked against evidence.
- Load-bearing: passes — The paper’s conclusions depend on it.
- Based: passes — The paper gives a reason why it holds.
A conclusion the paper draws from its own results — it cites no other source for it, so there is none to check. The quote shows it is in the paper.
In the paper
“The absence of a standardized, cryptographically resilient provenance model creates operational challenges for organizations deploying AI systems in high-assurance environments.”
plainstates a cause- Asserted: passes — The authors put it forward as true.
- Checkable: passes — It can be checked against evidence.
- Load-bearing: passes — The paper’s conclusions depend on it.
- Based: unclear — It is unclear why the paper says it holds.
36 more claims
Create a free account to see all 135 claims and to verify the 40 references still open.
- It may state two separate claims as one. 51
- Probably not a substantial claim. 72
Self-check
Percentages, sums and reported differences re-computed from the paper’s own numbers.
Each contradiction between two passages listed here was confirmed by an independent check.
Rated minor: it would not touch a result — a value no result rests on, a p-value that stays on the same side of the significance level, or a conflict that is clerical, resolved elsewhere in the paper, or outside its own findings.
In the paper
§ 6.1. Pipeline Overview · p. 17
“Classical Signing (RSA) in parallel → Hybrid Signature Bundle”
§ 6.1.3. Stage 3-Signing · p. 18
“During the transition period, hybrid bundles combining a classical ECDSA signature with the appropriate PQC signature are generated and stored with the artifact.”
Citations
Cites
The PRISMA 2020 statement: An updated guideline for reporting systematic reviews
Page, M. J., McKenzie, J. E., Bossuyt, P. M., Boutron, I., Hoffmann, T. C., Mulrow, C. D., Shamseer, L., Tetzlaff, J. M., Akl, E. A., Brennan, S. E., et al. (2021) · BMJ · doi:10.1136/bmj.n71 · Cited by 108,283
Design Science in Information Systems Research1
Hevner, A. R., March, S. T., Park, J., & Ram, S. (2004) · MIS Quarterly · doi:10.2307/25148625 · Cited by 10,707
A Design Science Research Methodology for Information Systems Research
Peffers, K., Tuunanen, T., Rothenberger, M. A., & Chatterjee, S. (2007) · Journal of Management Information Systems · doi:10.2753/mis0742-1222240302 · Cited by 7,832
BadNets: Evaluating Backdooring Attacks on Deep Neural Networks
Gu, T., Liu, K., Dolan-Gavitt, B., & Garg, S. (2019) · IEEE Access · doi:10.1109/access.2019.2909068 · Cited by 1,238
Rose, S., Borchert, O., Mitchell, S., & Connelly, S. (2020) · doi:10.6028/nist.sp.800-207 · Cited by 1,174
Security and Privacy Controls for Information Systems and Organizations
Joint Task Force Interagency Working Group (2020) · doi:10.6028/nist.sp.800-53r5 · Cited by 419
Dataset Security for Machine Learning: Data Poisoning, Backdoor Attacks, and Defenses
Goldblum, M., Tsipras, D., Xie, C., Chen, X., Schwarzschild, A., Song, D., Madry, A., Li, B., & Goldstein, T. (2023) · IEEE Transactions on Pattern Analysis and Machine Intelligence · doi:10.1109/tpami.2022.3162397 · Cited by 362
Artificial Intelligence Risk Management Framework (AI RMF 1.0) - Japanese translation
NIST, G. M. (2024) · doi:10.6028/nist.ai.100-1.jpn · Cited by 269
Backstabber’s Knife Collection: A Review of Open Source Software Supply Chain Attacks
Ohm, M., Plate, H., Sykosch, A., & Meier, M. (2020) · Lecture Notes in Computer Science · doi:10.1007/978-3-030-52683-2_2 · Cited by 259
BackdoorBench: A Comprehensive Benchmark of Backdoor Learning
Wu, B., Chen, H., Zhang, M., Zhu, Z., Wei, S., Yuan, D., & Shen, C. (2022) · neural information processing systems · doi:10.52202/068431-0766 · Cited by 232
Research works printed but not linked: 2
- NIST (2024). Artificial Intelligence Risk Management Framework: Generative Artificial Intelligence Profile. · DOI · Paper page
- Li, Y., Lyu, X., Koren, N., Lyu, L., Li, B., and Ma, X. (2021). Anti-Backdoor Learning: Training Clean Models on Poisoned Data. Proceedings of the 35th Conference on Neural Information Processing Systems (NeurIPS 2021), Curran Associates.
Cited by
Post-Quantum Cryptography Migration for Agentic AI Systems
Campbell, R. (2026) · Computers · doi:10.3390/computers15070434 · Cited by 5
Self-citationCampbell, R. (2026) · Computers · doi:10.3390/computers15100659 · Cited by 1
Self-citation
Metadata
- Authors
- Campbell, R.
- Affiliations
- Independent Researcher, Upper Marlboro, MD 20774, USA
- Corresponding author
- Campbell, Robert
- ORCID
- Campbell, Robert
- Venue
- Systems
- Publisher
- MDPI AG
- Published
- 2026-05-21
- Volume
- 14
- Issue
- 5
- Pages
- 593
- ISSN
- 2079-8954 (ISSN-L)
- OpenAlex
- W7162027274
- Cited by
- 3
- References
- 16
- Citations by year
- 2026: 3
- FWCI
- 10.25
- Type
- journal article
- Language
- en
- Open access
- gold
- doi.org
- Licence
- Locations
- Systems (publishedVersion, cc-by)
- Preprints.org (acceptedVersion, cc-by)
- DOAJ (DOAJ: Directory of Open Access Journals) (submittedVersion, cc-by-sa)
- Topics
- Physical Sciences › Computer Science › Artificial Intelligence › Adversarial Robustness in Machine Learning
- Physical Sciences › Computer Science › Computer Networks and Communications › Cryptography and Data Security
- Physical Sciences › Computer Science › Information Systems › Blockchain Technology Applications and Security
- Keywords
- model provenance, post-quantum cryptography, digital signatures, ML-DSA, model poisoning, AI governance, root of trust
About these checks
Registry, retraction, Cites and statistics results are computed, not judged, with one exception: a printed reference that no index lists is read once by a model before it counts as a citation error. Claims are extracted by a model and kept only when their quote appears verbatim in the text; each cited source is then checked, with a model’s help, for whether it says what the claim relies on. Those verdicts can be wrong — read the claim’s quote and the cited source before relying on one.
The claim verdicts this report counts are experimental: their accuracy is not yet measured.
A language model drew the claims from the paper on 2026-10-29, under version 3 of the claim definition (model deepseek-v4.1-flash, prompt extract-v3).
A second model labelled each claim on 2026-10-29 (model jev-1.13-20260917, definition revision gold-guideline-v0.2).
Cited sources are checked with a language model’s help (gpt-6-sol).
Statistics and arithmetic are recomputed, not judged.
Confidence bands are shown only for fields calibrated on our annotated test set. None is calibrated yet, so no band is shown.
What counts as a claim ›How we measure confidence ›How we check this ›
Are you an author, or do you see a problem with this page? Request a review
How to cite
Cite CoChat for the checks on this page, and the paper for its own findings.
- This page
- CoChat. (2026). Citation check: AI Supply Chain Security: MBOM-PQC Provenance, PQC Attestation, and a Maturity Model for Quantum-Resistant Assurance. https://app.cochat.ai/papers/10.3390/systems14050593
- The paper
- Campbell, R. (2026). AI Supply Chain Security: MBOM-PQC Provenance, PQC Attestation, and a Maturity Model for Quantum-Resistant Assurance. Systems. https://doi.org/10.3390/systems14050593
Data sources
Bibliographic data: Semantic Scholar, OpenAlex, Crossref.
Contains information from the Semantic Scholar Academic Graph Datasets, made available under the ODC Attribution License (ODC-BY 1.0).
Semantic Scholar’s data is described in: Kinney, R., et al. (2023). The Semantic Scholar Open Data Platform. arXiv:2301.10140.