Primary Sources and Open Checks
The official texts this reference relies on (EU legal instruments with ELI links, DORA Level 2 acts, Commission and ESA publications, national law, standards, vendor documentation and the specifications hosted on this site), each with its identifier, the version or date it was read, its verification status and the pages that use it, followed by a consolidated register of the open fact-checks that still stand between a source claim and an official text.
- Version
- 0.1.0
- Owner
- DEIXEI
- Last reviewed
- Reading time
- 28 min
For Cto · Ai Architect · Platform Engineer · Risk And Assurance · Compliance · Legal And Compliance · Internal Audit
Sources and details
- Section
- Reference
- Source works
- Book 1 · Book 3 · Research Series · Supporting Tools
- Source chapters
- 6, 7
- Source appendices
- 01:C, 04:A, 04:I
- Tool specifications
- audit_evidence_schema@0.1.0, cab_decision_record@0.1.0, deployer_manifest@0.1.0
- Refresh trigger
- quarterly
This page is the bibliography of official texts behind the AI Delivery Factory reference, and the register of what has not yet been checked against them. It answers three questions a reviewer, an auditor or a supervisor-facing team asks before relying on a page:
- Which official text does this statement rest on, and which version was read?
- Has anyone on this reference actually read that text, or is the statement only as a book cites it?
- What is still open, who closes it, and where?
The books, the research papers and the Supporting Tools are the reference’s secondary sources. They supply the method, the models and most of the numbers, and each page names them in its own source notes. This page lists only the texts those sources point at: legislation, delegated and implementing acts, supervisory publications, standards, vendor documentation, and the specifications this site publishes.
It was compiled on 14 November 2026 from two things: every sourceUrls entry across the reference, and every [FACT-CHECK: …] marker in page text (347 markers on 37 pages at that date). A recount during verification on 15 November 2026 found 355 markers on 43 pages, because pages written in the current wave had been added; those pages are now placed in the groups below. A statement that carries a marker is an open check, not a source, and it appears in the open-checks register below, not in the source tables.
How to use this page
| You need to… | Do this |
|---|---|
| Cite an obligation externally | Use only a row marked Verified below, then read the text yourself at the official location. A book citation is not an official source. |
| Plan against a date | Go to the regulatory calendar. This page deliberately does not restate application, publication or deadline dates. |
| Map an article to a control | Go to the regulatory map. This page names instruments; the map reads them. |
| Configure a vendor product | Read the vendor row here for what was verified and when, then check the open-checks group for that vendor before rollout. |
| Close a check | Follow the closing procedure. |
Status labels
| Label | Meaning | May a page state it as fact? |
|---|---|---|
| Verified | Read against the official location on the date given, by the page that cites it | Yes, with the date |
| Verified (secondary host) | Read from an official consolidation or official publisher other than EUR-Lex (the Commission’s AI Act Service Desk; the Official Journal text hosted outside EUR-Lex) because EUR-Lex refused automated retrieval during the November 2026 check | Yes, with the host and date; confirm on EUR-Lex before external citation |
| Checked (unofficial copy) | Read against an unofficial reproduction of the official text | No. Treat as cited until read on EUR-Lex |
| Cited | Named by a book or paper; the official text was not read for this reference | No. Attribute it (“per Book 1 Ch 6”) and keep the marker |
| Vendor-dated | A vendor’s own documentation, read on a date; vendor documentation is not the contract | Yes, as “the vendor’s documentation states, as of …”; the executed agreement governs |
EU legal instruments
Level 1: regulations the reference is built on
| Instrument | Identifier and version read | Official location | Status | Used on |
|---|---|---|---|---|
| Artificial Intelligence Act | Regulation (EU) 2024/1689, as amended by Regulation (EU) 2026/1744 (the AI Omnibus). Articles 14, 26, 27, 49, 50, 73 and 113 read in the Service Desk consolidated text as at 27 July 2026, accessed 15 Nov 2026 | ELI: eur-lex.europa.eu/eli/reg/2024/1689/oj/eng; consolidated articles at the AI Act Service Desk | Articles listed: Verified (secondary host). All other articles: Cited | Regulatory map, regulatory calendar, control map, risk tiers, model governance, audit evidence, incident response, data classes and lanes, authority process |
| AI Omnibus (amending act) | Regulation (EU) 2026/1744. The OJ reference was confirmed through the Commission’s AI Act policy page (last updated 19 Oct 2026); the regulation number is as Book 2 Ch 26, Book 09 Ch 1 and Research Paper 1 cite it | Commission AI Act policy page; ELI not yet recorded | OJ reference: Verified (secondary host). Act date, OJ date and the other reported amendments: Cited | Regulatory calendar, regulatory map |
| DORA (Digital Operational Resilience Act) | Regulation (EU) 2022/2554. Application date confirmed on the EIOPA DORA page, 15 Nov 2026 | ELI: eur-lex.europa.eu/eli/reg/2022/2554/oj/eng; EIOPA DORA page | Application date: Verified. Article-level statements: Cited unless the regulatory map marks them verified | Regulatory map, regulatory calendar, control map, vendor concentration and exit, deployment routes, incident response, change governance, release engineering, evaluation and resilience and most build pages through the CIF tag |
| GDPR | Regulation (EU) 2016/679. Articles 22, 28 and the processor and sub-processor concepts are used; no article was re-read for this reference | ELI not yet recorded | Cited | Regulatory map, data classes and lanes, deployment routes, risk tiers, incident response, identity and secrets, coding-agent workstation |
The control map cites the AI Act by a shorter ELI (/eli/reg/2024/1689) accessed 14 Nov 2026. It resolves to the same act; use the /oj/eng form above in new citations so every page points at one address.
Level 2: DORA delegated and implementing acts
Regulatory technical standards (RTS) are adopted as Commission delegated regulations and implementing technical standards (ITS) as Commission implementing regulations. The reference cites them by their short number, as the books do.
| Instrument | What the reference uses it for | Official location | Status | Used on |
|---|---|---|---|---|
| RTS 2025/301: content and time limits for major-incident notifications and reports | The 4 h / 24 h / 72 h / one-month clocks and when they start (Art 5) | ELI: eur-lex.europa.eu/eli/reg_del/2025/301/oj/eng; OJ text read 15 Nov 2026 | Art 5: Verified | Regulatory map (clocks), regulatory calendar, incident response |
| RTS 2024/1774: ICT risk-management framework | Art 17 change management as the evidence frame for change records; the logging article | ELI not yet recorded | Art 17(1): Checked (unofficial copy). Logging article number and OJ date: Cited | Regulatory map, regulatory calendar, audit evidence, change governance |
| RTS 2024/1772: classification of ICT-related incidents | Art 8 major-incident test | ELI not yet recorded | Art 8: Checked (unofficial copy). OJ reference: Cited | Regulatory map, regulatory calendar, incident response |
| RTS 2024/1773: policy on ICT services supporting critical or important functions provided by third parties | Supplier-policy rows; the policy count Book 1 gives | ELI not yet recorded | Cited | Regulatory map, regulatory calendar, vendor concentration and exit |
| ITS 2024/2956: Register of Information templates | Which template carries the arrangement and service rows | ELI not yet recorded | Cited (Book 1 names template B_05.02) | Regulatory map (Register of Information), regulatory calendar, vendor concentration and exit |
| ITS 2025/302: incident-reporting templates | Field list for the initial notification | ELI not yet recorded | Cited | Regulatory calendar, incident response |
| RTS 2025/532: subcontracting of ICT services supporting critical or important functions | Supplier and sub-processor rows | ELI not yet recorded | Cited | Regulatory map, regulatory calendar |
| RTS 2025/1190: threat-led penetration testing | TLPT cadence, tester rotation and entity scope; agents in TLPT scope | ELI not yet recorded | Cited | Regulatory map (TLPT), regulatory calendar, evaluation and resilience |
Recommendation. The DORA Level 2 acts are the largest block of unverified law in the reference, and three of them drive evidence the factory produces every day (change records, incident classification, the Register of Information). Close OC-01 before any of those pages moves out of draft.
Sectoral law cited by the reference
These instruments bind regardless of AI. The reference cites them through Book 3 Appendix I §8 (NIS2 through §9) and Book 1 Ch 6–7. None was re-read for this reference; all are Cited. Identifiers are deliberately not given until they are read on EUR-Lex (OC-09).
| Instrument | Provisions the reference uses | Used on |
|---|---|---|
| MiFID II and RTS 6 | Arts 16, 17 (algorithmic trading, with the RTS 6 self-assessment), 24 and 25 | Regulatory map |
| CRR3 and CRD VI | Internal-model requirements, read with the ECB Guide to Internal Models | Regulatory map, model governance |
| Solvency II | Art 41 (governance); Arts 112–127 (internal models) | Regulatory map, authority process, implementation roadmap |
| Insurance Distribution Directive (IDD) | Art 25 product oversight and governance | Regulatory map |
| PRIIPs Regulation | Manufacturer accuracy duty for key information documents | Regulatory map |
| NIS2 | Incident-reporting overlap with DORA | Regulatory map, incident response |
National law and case law
| Instrument | Version read | Official location | Status | Used on |
|---|---|---|---|---|
| Italy, Legge 23 settembre 2025, n. 132 (AI law) | Gazzetta Ufficiale, Serie Generale n. 223, accessed 15 Nov 2026 | ELI: gazzettaufficiale.it/eli/id/2025/09/25/25G00143/sg | Adoption, publication and entry-into-force dates: Verified. Content as described by Book 1 Ch 6–7; delegated decrees: Cited | Regulatory calendar (with the correction of Book 1 App C), regulatory map |
| Germany (KI-MIG), Netherlands (Uitvoeringswet AI-verordening), Spain (anteproyecto), France, Belgium | Draft or proposal status as of 30 May 2026 per Book 1 Ch 7 | Bundesgesetzblatt, Staatscourant, BOE, Légifrance, Moniteur belge | Cited; none is yet a source | Regulatory calendar |
| Switzerland: revFADP (Arts 9(1), 21, 22, 24); FINMA Guidance 08/2024; FINMA Circulars 2023/1 and 2018/3; Swiss Bankers Association Cloud Guidelines, 3rd edition | Circular 2018/3 margin numbers read by Research Paper 9 against the circular as amended 4 November 2020; the rest as Book 3 App I cites them | FINMA; fedlex; SBA | Cited (Paper 9’s reading of 2018/3 is a secondary check, not a check by this reference) | Regulatory map (Switzerland), data classes and lanes, deployment routes, control map |
| United Kingdom: Data (Use and Access) Act 2025 (substitution of UK GDPR Art 22) | As of 10 Nov 2026 per Research Paper 2 | legislation.gov.uk | Cited | Regulatory calendar, regulatory map |
| CJEU, SCHUFA judgment (C-634/21) | Case reference and date as Book 3 cites them | curia.europa.eu | Cited | Regulatory map (GDPR Art 22), risk tiers |
Commission and supervisory publications
None of these is binding law. The regulatory map labels them Guidance. Publication dates are on the regulatory calendar.
| Publication | Reference or version | Official location | Status | Used on |
|---|---|---|---|---|
| Commission AI Act Service Desk: implementation timeline; consolidated articles | Consolidated as at 27 July 2026; accessed 15 Nov 2026 | Timeline; articles 14, 26, 27, 49, 50, 73, 113 | Verified | Regulatory map, regulatory calendar |
| Commission AI Act policy page | Last updated 19 Oct 2026; accessed 15 Nov 2026 | digital-strategy.ec.europa.eu | Verified | Regulatory calendar (Omnibus facts) |
| ESAs, first list of designated critical ICT third-party providers | ESMA press release; accessed 15 Nov 2026 | ESMA press release | Publication: Verified. List content (which providers are on it, and that no model provider is): Cited from Book 3 | Regulatory calendar, regulatory map, deployment routes, vendor concentration and exit, incident response |
| EIOPA, Opinion on AI governance and risk management | EIOPA-BoS-25-360; date and reference confirmed 15 Nov 2026 | EIOPA Opinion | Date and reference: Verified. The seven principles and four assessment dimensions: Cited | Regulatory map (sectoral layer), regulatory calendar, implementation roadmap |
| EBA factsheet on AI Act implications for the banking sector | Date confirmed via the EBA listing on 15 Nov 2026; canonical title differs between the PDF and Book 1 | EBA publications | Date: Verified. Title and content: Cited | Regulatory map, regulatory calendar, model governance |
| ECB Guide to Internal Models (revised) | As Book 3 App I cites it | ECB Banking Supervision | Cited | Regulatory map, regulatory calendar, model governance |
| ESMA statement on AI in retail investment services; ESMA supervisory briefing on algorithmic trading | As Book 3 App I cites them | ESMA | Cited | Regulatory map, regulatory calendar |
| BaFin guidance on ICT risks in the use of AI at financial entities | As Book 3 App I cites it | BaFin | Cited | Regulatory calendar, regulatory map |
| Commission guidelines on high-risk classification (AI Act Art 6(5)); guidance on Art 73 serious-incident reporting; codes of practice under Art 50(7) | Delayed, unverified or pending, per the calendar | Commission; Service Desk | Not yet sources | Regulatory calendar, risk tiers, incident response |
Standards, protocols and threat catalogues
Voluntary unless the firm adopts them or another instrument makes them binding. The control map records, for each, the edition, the adoption basis and the practices the firm chose to implement. No standard below was read against its official text for this reference. Where the books disagree on a version, the page that uses it says which it follows.
| Standard or catalogue | Edition the sources cite | Publisher | Used on |
|---|---|---|---|
| ISO/IEC 42001 (AI management system) | 2023, with Annex A control anchors | ISO | Control map, regulatory map, regulatory calendar, deployment routes, evaluation and resilience |
| ISO/IEC 23894 (AI risk management); ISO/IEC 27001 and 27701 | Not stated in the sources | ISO | Control map, regulatory map, MCP and connector broker |
| NIST AI RMF; NIST AI 600-1 (Generative AI Profile); NIST AI 100-2 (adversarial ML); NIST SP 800-218A | AI RMF 1.0 | NIST | Control map, regulatory map, evaluation and resilience |
| OWASP Top 10 for Agentic Applications (ASI01–ASI10); OWASP Top 10 for LLM Applications | Agentic list v1.0 per Book 1 App C; the books differ slightly on version labels and titles | OWASP GenAI Security Project (genai.owasp.org) | Evaluation and resilience, control map, incident response, risk tiers, MCP and connector broker, and others |
| MITRE ATLAS | v5.1.0 per Book 3 App I | MITRE (atlas.mitre.org) | Evaluation and resilience, control map, risk tiers, MCP and connector broker |
| CIS Controls; CIS benchmarks | Controls v8.1 (18 controls, 153 safeguards); benchmark v5.0.0 as the source cites it | Center for Internet Security | Control map |
| ITIL; COBIT | ITIL 5 as Book 1 Ch 5 cites it; ISACA’s 2025 paper on COBIT for AI governance | PeopleCert / AXELOS; ISACA | Control map, readiness diagnostic, implementation roadmap |
| CEN-CENELEC harmonised standards under standardisation request M/593 (including prEN 18286, QMS) | Enquiry stage per Book 1 App C | CEN-CENELEC; citation in the OJ | Regulatory calendar |
| Model Context Protocol | Revision 2026-07-28 (previous 2025-11-25); Book 3 App C was written against 2025-11-25 | MCP specification site | MCP and connector broker, change governance |
| Agent2Agent (A2A) protocol | Latest released specification; version not pinned in the sources | A2A project | MCP and connector broker, BOM and recipes, evaluation and resilience |
| in-toto Statement | v1 (_type: https://in-toto.io/Statement/v1), used as the AES envelope; the spec does not redefine it | in-toto project | Audit evidence, schemas and tools |
| JSON Schema | Draft 2020-12 (the AES schema’s dialect) | json-schema.org | Schemas and tools, audit evidence, release engineering |
Vendor documentation
The site is not vendor-neutral. Vendor facts sit on the topic pages, dated, and the executed agreement always overrides public documentation. Only the rows below were read directly by a page of this reference; everything else vendor-specific is carried as an open check (OC-18 to OC-27).
Read directly
| Vendor page | Read on | Status | Used on |
|---|---|---|---|
| Anthropic, API and data retention | 15 Nov 2026 | Vendor-dated | Deployment routes (ZDR by route, features that leave ZDR, HIPAA) |
| Anthropic, Data residency | 15 Nov 2026 | Vendor-dated | Deployment routes (residency checks) |
| Anthropic, Claude in Microsoft Foundry | 15 Nov 2026 | Vendor-dated | Deployment routes (Anthropic as processor on Foundry; hosting options) |
| Anthropic, Claude Platform on AWS | 15 Nov 2026 | Vendor-dated | Deployment routes (inference location; MCP tunnels; separate organisation) |
| Anthropic, Claude models on Google Cloud, Anthropic managed | 15 Nov 2026 | Vendor-dated | Deployment routes (an offering absent from Book 3) |
| Anthropic, Claude Agent SDK: hosting, session storage, permissions | 15 Nov 2026 | Vendor-dated | Deployment routes (SDK as a process; session-store mirroring; permission pipeline) |
| Microsoft Learn, Azure Blueprints retirement | Page dated 30 Oct 2026; checked 14 Nov 2026 | Vendor-dated | Landing zone (retirement timeline and successors) |
| Microsoft Learn, Power Platform environments overview and Secure the default environment | Cited by Book 2 Ch 21 and Book 09 Ch 10 (November 2026); the URLs are held on the page | Cited (the page reports the books’ reading; the landing-zone page keeps a check open) | Application foundation, landing zone |
What the direct reading changed. Reading the vendor text on 15 November 2026 overturned several Book 3 statements, and the deployment-routes page follows the vendor text on each. Among them: who processes on Foundry (where Book 3 itself is split between appendices); inference location on Claude Platform on AWS; HIPAA eligibility per route; whether the local session file stays authoritative once a session store is in use; computer-use eligibility; and whether Foundry publishes feature-limitation and networking guidance. The difference table is on deployment routes. The glossary’s hosting-route crosswalk now follows the same reading.
Secondary vendor snapshots
Two research papers carry dated vendor-documentation snapshots that pages use with attribution. They are secondary sources, not verified primary ones:
- Research Paper 9 §8.2 (accessed 10 November 2026): training, retention and processor statements for Microsoft, Amazon Bedrock, Google Cloud and Anthropic. Used on data classes and lanes and model gateway. See /research/data-classes-processing-lanes-and-roles/#sec-8-2.
- Research Paper 3 (verified 10 November 2026): Claude Code and Foundry statements. Used on deployment routes and coding-agent workstation. See /research/securing-claude-code-in-the-enterprise/.
Specifications published on this site
These are DEIXEI’s own normative files, versioned on the Supporting Tools cadence. They are primary for the fields they define. Schemas and tools explains what each requires and where the source specification contradicts itself.
| Specification | Version | Location | Used on |
|---|---|---|---|
| Audit Evidence Schema (AES predicate) | v0.1.0 | /aes/v0.1.0/audit_evidence_schema.json | Audit evidence, accountability, schemas and tools, enterprise control plane, agent lifecycle, control map, agent-driven departments |
| CAB change (input) | v0.1.0 | /aes/v0.1.0/cab_change.json | Change governance, accountability, schemas and tools |
| CAB decision (output) | v0.1.0 | /aes/v0.1.0/cab_decision.json | Change governance, schemas and tools |
| Deployer Manifest | v0.1.0 | /aes/v0.1.0/deployer_manifest.json | Release engineering, accountability, schemas and tools |
Open-checks register
Each group below consolidates the [FACT-CHECK: …] markers on the named pages. A group is closed only when every marker in it has been resolved on its page. Priority is this page’s recommendation. P1 blocks an external citation, a supervisory answer or a regulated go-live. P2 blocks a configuration rollout or a contract row. P3 is context that should be confirmed before publication. Owner names the role that should close the group, also a recommendation.
Law and supervisory publications
Owner for every row: the regulatory-calendar owner in Legal and Compliance. The calendar is the canonical place to record any date the check produces.
| ID | Open check | Where to check | Pages carrying it | Priority |
|---|---|---|---|---|
| OC-01 | ELI links, OJ references and article pinpoints for the DORA Level 2 acts: RTS 2024/1774 (which article is the logging article; Art 17 read only on an unofficial copy), RTS 2024/1772 (Art 8), RTS 2024/1773 (the policy count Book 1 gives), ITS 2025/302 (template fields), ITS 2024/2956 (which template carries arrangement and service rows), RTS 2025/532, RTS 2025/1190 (cadence, tester rotation, entity scope) | EUR-Lex | Regulatory map, regulatory calendar, incident response, audit evidence | P1 |
| OC-02 | The “five-year floor” Book 1 App B attributes to DORA Art 9. That article is not obviously a retention-period provision | DORA and RTS 2024/1774 on EUR-Lex | Audit evidence | P1 |
| OC-03 | DORA Art 30(2) versus 30(3) (which contract provisions apply to all ICT arrangements and which only to those supporting critical or important functions); the basis of the first-TLPT deadline; the Register of Information submission window this year | EUR-Lex; the firm’s competent authority | Regulatory map, regulatory calendar, vendor concentration and exit | P1 |
| OC-04 | AI Omnibus (Regulation (EU) 2026/1744): date of the act, OJ publication date, ELI, and each of the other reported changes (Art 4 AI literacy, bias-detection necessity test, small mid-cap relief, financial-sector competence) | EUR-Lex | Regulatory calendar, regulatory map | P1 |
| OC-05 | AI Act provisions not yet re-read: Art 111(3) (legacy GPAI models); Art 86 scope after the amended Art 113; Art 49(2) as consolidated; Annex III exclusions for AML and trading models; point lettering in the reproduced Art 14(4) elements | EUR-Lex; Service Desk | Regulatory map, regulatory calendar, risk tiers, model governance | P1 |
| OC-06 | Pending Commission deliverables: Art 6(5) high-risk classification guidelines; Art 73(7) serious-incident guidance; Art 50(7) codes of practice | Commission; Service Desk | Regulatory calendar, risk tiers, incident response | P1 |
| OC-07 | Content of the first CTPP designation list (Book 3: nineteen providers, including AWS, Microsoft and Google Cloud; neither Anthropic nor OpenAI) and any later round | ESAs list PDF; ESAs Joint Committee | Regulatory calendar, regulatory map, deployment routes, vendor concentration and exit, incident response | P1 |
| OC-08 | DORA incident report field list against the RTS and ITS templates (Book 3’s extract is illustrative); the incident-criteria count where Book 1, Book 3 App I and App F differ | EUR-Lex | Incident response, regulatory map | P1 |
| OC-09 | Identifiers and consolidated versions of the sectoral instruments (MiFID II and RTS 6, CRR3/CRD VI, Solvency II, IDD, PRIIPs, NIS2) and of GDPR | EUR-Lex | Regulatory map, model governance, authority process | P2 |
| OC-10 | Supervisory publications: EBA factsheet canonical title; EIOPA Opinion’s seven principles and four dimensions; ECB Guide to Internal Models revision; ESMA retail statement and algorithmic-trading briefing; BaFin AI guidance; the “EBA factsheet of 21 November 2025” wording cited by Book 1 Ch 14; the BaFin, CSSF and ESRB statements Book 3 cites on model change; EBA and EIOPA follow-up work | Issuing authorities’ sites | Regulatory map, regulatory calendar, model governance, evaluation and resilience | P2 |
| OC-11 | Switzerland: FINMA Guidance 08/2024 content; SBA Cloud Guidelines, 3rd edition; revFADP articles | FINMA; fedlex; SBA | Regulatory map, regulatory calendar | P2 |
| OC-12 | National layer: which Italian legislative decrees under Art 24 of Law 132/2025 were adopted and whether the delegation period held; draft status in Germany, the Netherlands, Spain, France and Belgium; UK Data (Use and Access) Act commencement | Gazzetta Ufficiale and Normattiva; national journals; legislation.gov.uk | Regulatory calendar | P2 |
| OC-13 | CJEU SCHUFA judgment reference and date | curia.europa.eu | Regulatory map, risk tiers | P1 |
| OC-14 | Harmonised standards status (CEN-CENELEC, M/593, prEN 18286) and any OJ citation | CEN-CENELEC; OJ | Regulatory calendar | P2 |
Standards and protocols
| ID | Open check | Where to check | Pages carrying it | Owner | Priority |
|---|---|---|---|---|---|
| OC-15 | ISO/IEC 42001:2023 Annex A anchors. This reference reads A.10 as third-party relationships and puts event logging under A.6.2. Book 1 uses A.6.2.6 for audit logging in one chapter and for human oversight in another | The purchased standard | Control map | Head of model risk | P1 |
| OC-16 | OWASP Top 10 for Agentic Applications version, entry titles and v2.0 status; the OWASP incident tracker’s location, incident count and category distribution; the named public incidents | genai.owasp.org | Evaluation and resilience, incident response, regulatory calendar, recipe catalogue (ASI numbering) | CISO | P2 |
| OC-17 | MITRE ATLAS version; CIS benchmark version; ITIL 5 launch and module structure; any ISACA COBIT focus area for AI; the 80–90% framework-overlap figure, which is the author’s estimate with no primary source | atlas.mitre.org; cisecurity.org; PeopleCert; ISACA | Control map, regulatory calendar | CISO; head of operations | P3 |
| OC-18 | MCP revision 2026-07-28 key changes and deprecated features; elicitation; MCP Apps isolation and per-server disable; server-delivered skills extension; current A2A specification | MCP and A2A specification sites | MCP and connector broker, change governance | AI platform owner | P2 |
Vendor: Anthropic and Claude
Owner for every row: the AI platform owner, with the DPO for data-handling rows. As of November 2026 the books’ Claude statements date from May to August 2026; Claude Code settings in particular move with each release.
| ID | Open check | Where to check | Pages carrying it | Priority |
|---|---|---|---|---|
| OC-19 | Claude Code managed settings and versions: enforceAvailableModels, the allowManaged*Only lock flags, forceRemoteSettingsRefresh, disableAutoMode, processWrapper, claudeMdExcludes, allowedProviders, cleanupPeriodDays, the env block; the default model and permission mode by version; hook event names and blocking semantics; whether --bare skips managed hooks; whether a permissions.deny match emits a hook event; IDE policy names; authentication precedence; network host list | code.claude.com settings, managed-settings, hooks, network-config, authentication and model-configuration pages | Coding-agent workstation, model governance, release engineering, identity and secrets, landing zone, incident response, change governance, evaluation and resilience, bounded agent, schemas and tools, audit evidence, risk tiers | P2 |
| OC-20 | Data handling per feature: ZDR eligibility of the Files API, the Messages API mcp_servers connector and batch processing; prompt-caching TTL and ZDR; Compliance API coverage per surface (including whether the Cowork gap has closed); Managed Agents trace export and retention; local transcript persistence; extended-thinking trace returns | platform.claude.com data-retention pages; Claude Code ZDR page; admin documentation | Data classes and lanes, deployment routes, model gateway, audit evidence, bounded agent, vendor concentration and exit, research assistant | P1 |
| OC-21 | Route facts not re-read on 15 Nov 2026: Amazon Bedrock, Google-managed Google Cloud, Microsoft 365 Copilot (hosting, sub-processor listing and EU Data Boundary position), Managed Agents; per-model EU region availability (europe-west1, eu, Bedrock EU cross-region profile); Foundry EU data zone; any EU-resident Anthropic offering; SSO mediator and plan; contracting layers for the Anthropic-managed Google offering; MCP support per route | Vendor documentation per route; executed agreements | Deployment routes, cloud and hosting, data classes and lanes, vendor concentration and exit, landing zone, build-configure-consume, regulatory map, incident response, glossary (Bedrock and Google-managed processor), inspection case (Foundry region), research assistant (Claude Code routing to Foundry) | P1 |
| OC-22 | Contract terms that only the executed agreement settles: SLA, DPA notice periods, termination and data-return terms, MACC eligibility, current pricing | Executed MSA, DPA and SOW | Deployment routes, model gateway, vendor concentration and exit, build-configure-consume | P1 |
| OC-23 | Model line-up and lifecycle: current model IDs and aliases per route, alias resolution, whether a consolidated deprecation schedule is now published, retirement dates per route, the restriction Book 3 records on one model’s ZDR availability, the “best” alias behaviour | Anthropic models and deprecations pages; each cloud provider’s model lifecycle page | Model governance, release engineering, bounded agent, vendor concentration and exit, model gateway, schemas and tools, research assistant, inspection case | P2 |
| OC-24 | Cost and administration: prompt-caching multipliers, Batch API discount, apps-gateway capabilities, spend-limit semantics and default pricing tier, Enterprise admin cost controls, Analytics API scope, OTel metric names, contribution metrics (beta status, export, per-user attribution switch), the Value tab | Anthropic admin and API documentation | Model gateway, factory measurement, bounded agent, research assistant (prompt-caching terms) | P2 |
| OC-25 | Anthropic publications the books cite: the Usage Policy (prohibited and high-risk categories, revision date; anthropic.com/legal/aup); the constitution’s wording and value ordering (anthropic.com/constitution); “Zero risk isn’t the job: a CISO’s guide to agentic AI” (the four questions, the seven requirements, the least-agency wording); “Bringing code review to Claude Code” statistics; “Running an AI-native engineering org”; “Claude models explained”; an admin post on model defaults; the champion and communications kits; the Opus 4 to 4.5 upgrade account; the customer story and incident-reporting route; the trust-centre certification list; certification eligibility and pricing; Claude Security beta terms; Outcomes; Managed Agents scheduling and vault rotation; the finance plugins | anthropic.com, claude.com/blog, code.claude.com, trust.anthropic.com | Risk tiers, bounded agent, regulatory map, deployment routes, factory charter, enterprise control plane, model governance, incident response, accountability, agent-speed SDLC, agent lifecycle, implementation roadmap, evaluation and resilience, change governance, MCP and connector broker, research assistant (Usage Policy) | P2 (P1 for the Usage Policy rows, which are a contract boundary) |
Vendor: Microsoft, AWS and Google Cloud
Owner: the head of cloud for each provider, with the AI platform owner. Many of these were preview or announcement-stage when the books were written.
| ID | Open check | Where to check | Pages carrying it | Priority |
|---|---|---|---|---|
| OC-26 | Microsoft: Foundry versionUpgradeOption values and default; Foundry deployment types, data zones and the policy aliases that deny them; APIM MCP features and API Center MCP support; Entra Agent ID claims, Conditional Access status, sponsor and lifecycle workflows; Agent 365; Purview DLP on agent egress and Sentinel correlation; Purview Agent Framework SDK; Foundry PII detection; the Agent Governance Toolkit; Power Platform default-environment behaviour and Power Apps code apps; the FSI Landing Zone repository; immutable Blob retention semantics; product status per Book 1 Ch 5 | Microsoft Learn | Model gateway, release engineering, model governance, cloud and hosting, identity and secrets, MCP and connector broker, incident response, data classes and lanes, enterprise control plane, landing zone, application foundation, agent lifecycle, risk tiers, control map, BOM and recipes, implementation roadmap, schemas and tools, build-configure-consume, glossary and accountability (Entra sponsor workflows) | P2 |
| OC-27 | AWS: Bedrock model lifecycle states and end-of-life behaviour; IAM action names and the eu. inference profile; AgentCore Identity, Registry, Policy, Gateway and Runtime (Book 1 Ch 14 reports Registry and Policy GA in March 2026); Bedrock Guardrails; session-tag mapping to agent identity; CodeDeploy and AppConfig progressive delivery; CloudWatch signals; App Studio regions; AWS Proton support status; DynamoDB Global Tables consistency | AWS documentation | Release engineering, coding-agent workstation, model governance, enterprise control plane, implementation roadmap, running the factory, data classes and lanes, factory measurement, application foundation, build-configure-consume | P2 |
| OC-28 | Google Cloud: the agent-platform branding (Book 2 Ch 13 and others report Gemini Enterprise Agent Platform absorbing Vertex AI agent tooling in April 2026); model versions and retirement policy; Agent Identity and Agent Registry; Cloud DLP; Cloud Deploy; Firestore multi-region consistency; Application Design Center; Workload Identity Federation token lifetime | Google Cloud documentation | Model governance, release engineering, agent-driven departments, running the factory, data classes and lanes, enterprise control plane, application foundation, identity and secrets | P2 |
| OC-29 | Cross-provider: guardrail capability parity; agent-identity claim formats; evidence-store immutability guarantees; MCP connector token lifetime and revocation propagation; which instruction files (such as AGENTS.md) agents read by default, and AGENTS.md stewardship | Each provider’s documentation | Model gateway, audit evidence, agent lifecycle, incident response, BOM and recipes, application foundation, agent-speed SDLC; and the pages whose markers ask for every named cloud service to be re-checked: SDLC catalogue (including the starter kits), data, API and integration, agent readiness, choosing a product shape, policy, audit and actuals, workflow, recipe catalogue, agent-driven departments | P2 |
Closed by a direct reading, still marked elsewhere. The Azure Blueprints retirement timeline was verified on the landing-zone page on 14 November 2026. The markers on cloud and hosting, BOM and recipes and schemas and tools can be closed by pointing to that reading, after a re-read if the page has moved. The Foundry processor statement was verified on deployment routes on 15 November 2026. The build-configure-consume marker can be closed the same way.
Other third-party claims
| ID | Open check | Where to check | Pages carrying it | Owner | Priority |
|---|---|---|---|---|---|
| OC-30 | The Ponemon 2026 Cost of Insider Risks figure and its population | The report itself | Incident response | CISO | P3 |
| OC-31 | Market-data vendor licence terms on LLM or agent session use | Licence agreements | Risk tiers | Procurement | P2 |
| OC-32 | Works-council consultation as a go-live condition for per-user metrics in DE, AT and NL | Legal review | Factory measurement | Legal and Compliance | P1 |
| OC-33 | Supervisory expectation on production versus non-production exit rehearsal | Lead supervisor | Vendor concentration and exit | Second line | P2 |
Site specifications
| ID | Open check | Where to check | Pages carrying it | Owner | Priority |
|---|---|---|---|---|---|
| OC-34 | Licence for each hosted file under /aes/; publish a LICENSE notice before inviting external reuse | Site owner decision | Schemas and tools | Site owner | P1 |
| OC-35 | Field names and SLO targets in the running-the-factory tables against the current published Supporting Tools release | Supporting Tools release | Running the factory | AI platform owner | P2 |
| OC-36 | Book-internal question: whether Book 1 App F lists fewer standard-foundation components than the “eighteen” its prose and App D give | The author; Book 1 App F | SDLC catalogue | Site owner | P3 |
Close an open check
- Read the official text at the location in the register. A secondary host (a consolidation service, a vendor summary, a book) does not close a legal check. It can move the row to Verified (secondary host).
- Record what you read: the URL, the version or consolidation date, and the date read. For vendor pages, also record the product version the page describes.
- Fix the page that carries the marker. Replace the marker with the dated statement. Add the URL to that page’s
sourceUrls. If the official text contradicts a book, keep the book’s version visible in the page’s source-differences table, say which the page now follows, and log the book correction for the author. The books are never edited from this reference. - For a legal or supervisory item, run the calendar’s change-trigger procedure and record any date there, not here.
- Update this page. Move the item from the register into the right source table with its status and date. Delete the register row once no page still carries the marker.
- Re-date the page’s
updatedfield and, for vendor rows, set the next review against the page’svendor-releaseorquarterlytrigger.
Failure modes to avoid.
- Closing on the book. A check closed by re-reading the book that raised it is still open.
- Silent drift. A vendor row verified once and never re-read becomes wrong without anyone noticing. Vendor rows older than one quarter go back into the register.
- Date sprawl. A dated regulatory fact copied onto a build or operate page will drift from the calendar. Link instead.
- Treating documentation as contract. Processor, retention and notice statements go into a Register of Information entry or a contracting-matrix row only from the executed agreement. Book 3 App A §6 itself asks for the Foundry processor statement to be re-derived from current vendor text at every quarterly review; this page applies the same rule to every route (recommendation).
Acceptance checks for this page
- Every URL in any page’s
sourceUrlsappears in a source table here, with its status. As of 14 November 2026 the set is the 31 URLs in this page’s frontmatter, plus the control map’s short-form AI Act ELI, which is noted under the AI Act row. - Every page that carries a
[FACT-CHECK: …]marker appears in at least one register group. - No row here states an application, publication or deadline date for a legal instrument; those are on the calendar.
- Every row marked Verified names the date read and the host.
- Each register group names an owner and a priority, and P1 groups are reviewed before any page they touch moves from draft to review.
Source and status
This page’s synthesis: the status labels, the grouping of markers into the register, the priorities and the owners. Source claims are those recorded by the pages listed, with their dates. Verification dates come from the regulatory map and calendar (15 November 2026), deployment routes (15 November 2026), the landing-zone page (14 November 2026) and the control map (14 November 2026). This page did not read any official text itself. It compiles what the reference has read and what it has not.
Review trigger: quarterly, and whenever a page adds a sourceUrls entry or closes or adds a marker.
Source material
- Book 1 (Building Organizations for the Age of AI), Appendix C, The Stop-the-Press Calendar (as of 10 May 2026; the three-question discipline for a moving item)
- Book 1 (Building Organizations for the Age of AI), Chapter 6, DORA, the EU AI Act, and the Sectoral Layer (instrument list as cited, as of 15 May 2026)
- Book 1 (Building Organizations for the Age of AI), Chapter 7, The EU AI Act, Sectoral Guidance, and the National Layer (Draft 4, as of 30 May 2026)
- Book 3 (Deploying Claude Safely), Appendix A, Anthropic Deployment Options Matrix (revised 3 August 2026; §6 quarterly review checklist)
- Book 3 (Deploying Claude Safely), Appendix I, Regulatory Mapping (as of 14 May 2026; §11 maintenance triggers)
- Research Paper 3, Securing Claude Code in the Enterprise (vendor documentation as verified 10 November 2026)
- Research Paper 9, Data Classes, Processing Lanes and Controller/Processor Roles for AI (§8.2 vendor documentation snapshot, accessed 10 November 2026)
- Supporting Tools, spec/ (Audit Evidence Schema, CAB Decision Record and Deployer Manifest v0.1.0)
- European Commission, AI Act Service Desk, timeline and consolidated articles 14, 26, 27, 49, 50, 73 and 113 (consolidated as at 27 July 2026), accessed 15 November 2026, as recorded on the regulatory map and calendar
- Commission Delegated Regulation (EU) 2025/301, Official Journal text, read 15 November 2026, as recorded on the regulatory calendar
- Anthropic platform and Claude Code documentation pages listed on the deployment-routes page, read 15 November 2026
- Microsoft Learn, Azure Blueprints retirement (page dated 30 October 2026), checked 14 November 2026, as recorded on the landing-zone page
Related guidance
-
Regulatory Calendar for AI in EU Financial ServicesThe stop-the-press calendar rebuilt as a dated register: AI Act stages after the AI Omnibus, DORA and its RTS/ITS, CTPP designations, ESA and supervisor guidance, national laws such as Italy's Law 132/2025 with corrected dates, and harmonised standards, each with status, as-of date, official source and the reference pages it moves, plus the change-trigger procedure for when an item moves. -
Route Regulatory Questions to Obligations and EvidenceA dated, article-level reference that routes AI design questions to the EU AI Act, DORA and its RTS, GDPR and sectoral rules, and turns each obligation into a control, an owner and evidence: change records under RTS 2024/1774 Art 17, the 4 h / 72 h / one-month incident clocks, the CIF hinge and its five artefacts, the three-count Register of Information reconciliation, Art 14(4) oversight controls, the Art 26(5) discharge, the FRIA gate, Art 49 and Art 50, SCHUFA-grade review and supplier controls. -
Choose and Contract Claude Deployment RoutesDecide, per use-case class, which route carries Claude into production (Anthropic API, Claude Platform on AWS, Bedrock, Google Cloud, Microsoft Foundry, plus the Microsoft 365 Copilot surface); record who operates, bills and processes; size the diligence to the processor; check residency, zero data retention and HIPAA per feature; sign a contracting matrix row; and run Agent SDK or hosted agents with the controls the product does not supply. Provider-specific and dated. -
Data Classes, Processing Lanes and TokenisationClassify every model request by its data, route it only to a processing lane the class permits, keep derived artefacts inside their source class, and treat retention, roles and tokenisation as recorded, testable decisions. -
Vendor Concentration and ExitMeasure concentration on three axes against 70% and 85% discussion triggers, keep a tested successor for every material model usage, record eight contract-cleavage points per route, prove evidence portability with a 15-item checklist, and rehearse exit on one material use case every year with a filed close-out report. -
Build a Unified Control MapJoin the firm's AI and technology controls in one traceable operating view: rows are operating concerns, columns are frameworks read at their own altitude, every cell points to live evidence, and existing ITIL, COBIT and CIS artefacts are extended to agents with a five-question test instead of a parallel AI governance track. -
Schemas, Conformance Packs and the ops ToolchainWhat the versioned Supporting Tools schemas hosted under /aes/v0.1.0/ require, how to validate against them, where the source specification contradicts itself, and what the planned ops CLI, Claude plugins and service conventions define. -
Glossary and Vocabulary CrosswalkThe controlled vocabulary of the AI Delivery Factory reference, with a crosswalk for the places where the books, working papers and specifications use different words for the same thing.