Banking & Insurance
Legal Contract & Regulatory Clause Extraction
Limited Risk (Transparency)UnverifiedDiscuss / dispute
Parsing of ISDA Master Agreements, Credit Support Annexes and regulatory correspondence into structured collateral, termination and exposure data.
Consensus classification rationale: Contextual classification: pure extraction feeding a human credit-risk decision can stay limited-risk under the Art. 6(3) filter, but the moment the output drives an automated exposure or termination action it becomes preparatory to a high-risk decision. US books-and-records rules (SEC 17a-4, FINRA 4511) apply to the extracted records regardless of tier. a dissenting reading is recorded →
Indicative decision support, not legal advice. Risk classification depends on your concrete deployment context and can change with scope drift — validate the result with qualified counsel.
pricingPer completed contract redline / outcome-based contract resolution
oversightLegal counsel sign-off on escalated high-liability terms
Target market(s)European UnionUnited States (federal)change
Changes which instruments below count as in scope for this profile.
Target markets: European Union, United States (federal)
Regulatory footprint
5 instruments across 4 of 7 regulatory domains, plus 6 standards references- AI law1 instrument
- Data protection1 instrument
- Cyber & resilience1 instrument
- Online safety & platformsnone triggered
- Product safetynone triggered
- Financial services2 instruments
- Sector & employmentnone triggered
- Standards6 references
By jurisdiction
- EU3European UnionDORA, EU AI Act, GDPR
- US2United States (federal)FINRA Rule 4511 (General Books & Records), SEC Rule 17a-4 (US Records Retention)
The AI Act is one dimension of this footprint, not the whole of it — every domain above carries its own obligations and deadlines. See the instruments in the graph →
Confidence in this chain of evidenceConfidence: Check-worthy
The chain holds, but at least one hop rests on a secondary source, an ageing verification or a practice-derived step. Check the flagged hops before you rely on them.
Computed weakest-link over 23 evaluated hops across 1 target market: a chain is only as strong as its weakest step, so the band follows the worst hop rather than an average that would hide it. Five factors per hop — source tier, verification age, status certainty, community hardening, derivation kind — all read from graph data, never from a hand-set score.
Why this band8 factors lowered the band — each links to the claim behind it
- Source tier: FINRA Rule 4511 (General Books & Records) carries no resolvable citation — the claim is uncited. open node →
- Source tier: JTC 21 Technical Package (prEN 18228/18229/18281–83) rests on a secondary source (tracker or summary), not on the primary text. open node → primary source →
- Source tier: IEEE CertifAIEd™ carries no resolvable citation — the claim is uncited. open node →
- Source tier: prEN 18229-1 (Trustworthiness Framework, part 1) rests on a secondary source (tracker or summary), not on the primary text. open node → primary source →
- Status certainty: JTC 21 Technical Package (prEN 18228/18229/18281–83) is "draft", not settled in-force law. open node → primary source →
- Status certainty: prEN 18229-1 (Trustworthiness Framework, part 1) is "enquiry", not settled in-force law. open node → primary source →
- Verification age: FINRA Rule 4511 (General Books & Records) has no recorded verification date. open node →
- Verification age: IEEE CertifAIEd™ has no recorded verification date. open node →
Compliance brief
This use case is limited-risk under the EU AI Act (Limited Risk (Transparency)); transparency obligations apply.
What is owed
- Art. 50. Disclose AI interaction to natural persons; machine-readable marking of synthetic content; deepfake labelling; emotion-recognition disclosure.
- Art. 4. Providers and deployers must ensure sufficient AI literacy of staff dealing with AI systems.
- GDPR Art. 22. Right not to be subject to solely automated decisions with legal/similar effect; requires meaningful human involvement or explicit legal basis + safeguards.
- GDPR Art. 27. A controller or processor not established in the Union that falls within Art.
- GDPR Art. 17. Right to erasure collides with AI Act Art.
Dates that bind
- 2024-08-01 — AI Act enters into force. Regulation (EU) 2024/1689 in force; countdown for all staged obligations starts.
- 2025-02-02 — Prohibitions + AI literacy. Art. 5 prohibited practices ban applies (manipulation, social scoring, untargeted face scraping, workplace emotion recognition); Art. 4 AI literacy duty.
Maximum exposure
- EU AI Act: Tiered: €35m / 7% (prohibited practices); €15m / 3% (Art. 9–15 high-risk obligations incl. data governance, documentation, logging); €7.5m / 1% (Art. 99(5) — incorrect, incomplete or misleading information to notified bodies or national competent authorities)
- GDPR: Up to €20m or 4% of worldwide annual turnover
- SEC Rule 17a-4 (US Records Retention): SEC enforcement; multi-hundred-million-dollar off-channel/recordkeeping penalties are routine.
- FINRA Rule 4511 (General Books & Records): FINRA disciplinary action, fines and supervisory findings.
- DORA: Administrative penalties; periodic penalty payments for critical third parties
First five actions
- Confirm in writing whether this organisation builds/places the system on the market (provider) or only operates it (deployer), since the role is not yet established.
- Commission and confirm the Art. 50, Art. 4, GDPR Art. 22 obligations named above as active workstreams with an accountable owner.
- Stand up the named oversight design — Mode 1 — with a documented human-review procedure.
- Produce the technical documentation and evidence artefacts already mapped to this use case (Synthetic-Content Labelling / Watermarking, Interface Transparency & Content-Marking Layer, HITL Escalation Queue & Review UI) before they are requested.
- Put 2024-08-01 — AI Act enters into force — into the compliance calendar with an owner and lead time.
Terms used above: · · ·
Classification precedentcontested
Consensus reading: Limited Risk (Transparency) open in the graph →
Contextual classification: pure extraction feeding a human credit-risk decision can stay limited-risk under the Art. 6(3) filter, but the moment the output drives an automated exposure or termination action it becomes preparatory to a high-risk decision. US books-and-records rules (SEC 17a-4, FINRA 4511) apply to the extracted records regardless of tier.
What the reading rests on — the provisions this classification actually pulls in:
- Art. 50 — Transparency Duties
- Art. 4 — AI Literacy
- EU AI Act (Regulation (EU) 2024/1689)
- GDPR (Regulation (EU) 2016/679)
- SEC Rule 17a-4 (US Records Retention) (17 CFR 240.17a-4)
- FINRA Rule 4511 (General Books & Records) (FINRA Rule 4511)
- DORA (Regulation (EU) 2022/2554)
Dissenting views — recorded because the ambiguity is real, not because the consensus is weak:
- Limited risk (transparency duties only)editorial seed — dispute welcome
Pure clause extraction produces an input; the credit-risk decision stays with the human reviewer, so the system is not itself performing the Annex III 5(b) creditworthiness evaluation.
- High risk (Annex III 5(b))editorial seed — dispute welcome
Where the extracted clauses are the material basis of the lending decision and the reviewer rarely departs from them, the extraction is a component of the creditworthiness evaluation and inherits its class.
Disagree with either reading? propose a change with a source → · discuss this classification →
Baseline: of 100+, 40% were not definitively classifiable (18% clearly high-risk, 42% clearly low-risk). appliedAI Institute — AI Act risk classification of AI systems from a practical perspective
Applicable Regulations (5)
EU AI Act (Regulation (EU) 2024/1689)
unverified · verified 2026-08-12 source (as amended) EUR-LexAmended by Regulation (EU) 2026/1744. Verified 16 Aug 2026: the popular mirrors have not yet been updated — artificialintelligenceact.eu still serves the unamended 13 June 2024 text with no disclaimer, and the Commission's AI Act Service Desk pages still show pre-omnibus text with a visible omnibus disclaimer. Read the OJ or consolidated text on EUR-Lex. in force EU
Horizontal, risk-based product-safety law for AI systems and GPAI models. Extraterritorial market-place principle. Staged applicability 2025–2030 (Digital Omnibus: Art. 50 → 2 Aug 2026, Annex III → 2 Dec 2027, Annex I → 2 Aug 2028). (Digital Omnibus: Regulation (EU) 2026/1744, in force 27 July 2026).
Sanctions: Tiered: €35m / 7% (prohibited practices); €15m / 3% (Art. 9–15 high-risk obligations incl. data governance, documentation, logging); €7.5m / 1% (Art. 99(5) — incorrect, incomplete or misleading information to notified bodies or national competent authorities)
GDPR (Regulation (EU) 2016/679)
Applies unchanged next to the AI Act for all personal data in training, fine-tuning, RAG and inference. Key friction points: Art. 22 automated decisions, Art. 17 erasure vs. AI Act logging, Art. 35 DPIA.
Sanctions: Up to €20m or 4% of worldwide annual turnover
SEC Rule 17a-4 (US Records Retention) (17 CFR 240.17a-4)
Broker-dealer record retention: electronic records must be preserved in non-rewriteable, non-erasable (WORM) or audit-trail form, indexed and reproducible on demand. Retention is tiered, not flat — 17a-4(a) requires six years for blotters, ledgers and customer account records, while 17a-4(b) requires three years for the broader category of communications, trade confirmations and supporting records.
Sanctions: SEC enforcement; multi-hundred-million-dollar off-channel/recordkeeping penalties are routine.
FINRA Rule 4511 (General Books & Records) (FINRA Rule 4511)
unverified · no verification date in force US
Requires member firms to preserve books and records not otherwise specified for at least six years in a format compliant with SEA Rule 17a-4(f).
Sanctions: FINRA disciplinary action, fines and supervisory findings.
DORA (Regulation (EU) 2022/2554)
Digital operational resilience for the financial sector — in force and applying since 17 January 2025: ICT third-party risk (CTPP oversight), change management, resilience testing — applies to AI-based trading, credit and KYC systems. ICT third-party risk management extends to AI supply chains: model and dependency inventories (AIBOM/SBOM-style), contractual audit and exit rights, and concentration risk across model providers.
Sanctions: Administrative penalties; periodic penalty payments for critical third parties
Legal Obligations (11)
density
Art. 50 — Transparency Duties
Disclose AI interaction to natural persons; machine-readable marking of synthetic content; deepfake labelling; emotion-recognition disclosure.
in-force · verified 2026-08-16 source (as amended) EUR-Lexconvenience mirror — not updated artificialintelligenceact.euAmended by Regulation (EU) 2026/1744. Verified 16 Aug 2026: the popular mirrors have not yet been updated — artificialintelligenceact.eu still serves the unamended 13 June 2024 text with no disclaimer, and the Commission's AI Act Service Desk pages still show pre-omnibus text with a visible omnibus disclaimer. Read the OJ or consolidated text on EUR-Lex.
Art. 4 — AI Literacy
Providers and deployers must ensure sufficient AI literacy of staff dealing with AI systems. In force since 2 Feb 2025.
unverified · no verification date source (as amended) EUR-Lexconvenience mirror — not updated artificialintelligenceact.euAmended by Regulation (EU) 2026/1744. Verified 16 Aug 2026: the popular mirrors have not yet been updated — read the OJ or consolidated text on EUR-Lex.
GDPR Art. 22 — Automated Decisions
Right not to be subject to solely automated decisions with legal/similar effect; requires meaningful human involvement or explicit legal basis + safeguards.
unverified · no verification date read the article EUR-Lex
GDPR Art. 27 — EU Representative
A controller or processor not established in the Union that falls within Art. 3(2) scope (offering goods or services to, or monitoring the behaviour of, data subjects in the Union) must designate in writing a representative established in a Member State where the relevant data subjects are. The representative is mandated to be addressed by supervisory authorities and data subjects, in addition to or instead of the controller/processor, on all compliance issues — without prejudice to legal action against the controller/processor itself. Exempt: (a) occasional processing that does not involve large-scale special-category or criminal-conviction data and is unlikely to result in a risk to individuals, or (b) public authorities or bodies.
in-force · verified 2026-09-10 read the article EUR-Lex
GDPR Art. 17 — Erasure
Right to erasure collides with AI Act Art. 12 immutable logging — resolved architecturally via bitemporal data modelling + physical partition scrub.
unverified · no verification date read the article EUR-Lex
GDPR Art. 25 — Data Protection by Design
Privacy by design & default: minimisation, pseudonymisation, PII filters in pipelines and vector stores.
unverified · no verification date read the article EUR-Lex
GDPR Art. 35 — DPIA
Data-protection impact assessment for high-risk processing — pairs with AI Act fundamental-rights impact assessment (Art. 27) for public-facing high-risk systems.
unverified · no verification date read the article EUR-Lex
GDPR Art. 33/34 — Personal-Data Breach Notification
Notification of a personal-data breach to the supervisory authority and, where the risk to individuals is high, to the affected individuals themselves.
in-force · verified 2026-08-11 read the article EUR-Lex
GDPR Art. 32 — Security of Processing
Controller and processor implement technical and organisational measures appropriate to the risk, including pseudonymisation and encryption, and measures ensuring the ongoing confidentiality, integrity, availability and resilience of processing systems. Access to personal data by an unauthorised recipient — including one reached through a derived index such as a vector store — is the harm this article addresses.
in-force · verified 2026-08-17 read the article EUR-Lex
GDPR Art. 9 — Special Categories of Personal Data
Processing of health, biometric and other special-category data is prohibited unless one of the Art. 9(2) conditions applies; where it is permitted, the appropriate safeguards travel with it. This is the anchor for de-identification of clinical imaging and for the minimisation of health data in training and retrieval corpora.
in-force · verified 2026-08-17 read the article EUR-Lex
GDPR Art. 88 — Processing in the Employment Context
Opening clause: Member States may provide more specific rules for processing employees' personal data in the employment context, by law or by collective agreement, including suitable safeguards for human dignity, legitimate interests and fundamental rights, with particular regard to monitoring systems at the workplace. It is the bridge through which national employment rules — in Germany the BetrVG co-determination right and § 26 BDSG — govern workplace AI alongside the GDPR itself.
in-force · verified 2026-08-17 read the article EUR-Lex
Control Objectives (3)
obligation (article) → operationalized_by → control objective → satisfied_by → component/pattern; control objective → evidenced_by → evidence artifact
Art. 50
AI Interaction & Content Disclosure
Natural persons are informed they interact with an AI system, and generated/manipulated content carries both human-visible labels and machine-readable provenance (C2PA-class) that survives publication pipelines. Testable: disclosure presence across all interaction surfaces; watermark validity sampling post-publication; deepfake-path red-team (does stripped metadata get caught at the gate?).
satisfied by: Synthetic-Content Labelling / WatermarkingOutput Rails / Groundedness CheckInterface Transparency & Content-Marking Layer
evidenced by: Guardrail Telemetry & Sanitization Records
Art. 4
control layer: community mandate — propose objectives
GDPR Art. 22
control layer: community mandate — propose objectives
GDPR Art. 27
control layer: community mandate — propose objectives
GDPR Art. 17
control layer: community mandate — propose objectives
GDPR Art. 25
Vector & Chunk-Level Access Control
Practice-derived control objective (not named by any provision's own text): the requesting principal's read rights on every retrieved source segment are enforced before generation. Token- or claim-based ACL filtering is applied twice — at the chunker, which writes the source ACL into chunk metadata at ingestion, and at query time in the vector store, which filters candidates by the caller's entitlements — and a response-grounding check re-validates the caller's rights on each cited segment BEFORE the answer is composed. Testable: retrieval probe with a low-privilege principal against a restricted corpus; ACL drift reconciliation between source system and index; red-team reconstruction attempt from similarity results alone.
ISO/IEC 42001 clause A.7 (indicative)
evidenced by: Vector ACL Verification Report
GDPR Art. 35
control layer: community mandate — propose objectives
GDPR Art. 33/34
control layer: community mandate — propose objectives
GDPR Art. 32
Vector & Chunk-Level Access Control
Practice-derived control objective (not named by any provision's own text): the requesting principal's read rights on every retrieved source segment are enforced before generation. Token- or claim-based ACL filtering is applied twice — at the chunker, which writes the source ACL into chunk metadata at ingestion, and at query time in the vector store, which filters candidates by the caller's entitlements — and a response-grounding check re-validates the caller's rights on each cited segment BEFORE the answer is composed. Testable: retrieval probe with a low-privilege principal against a restricted corpus; ACL drift reconciliation between source system and index; red-team reconstruction attempt from similarity results alone.
ISO/IEC 42001 clause A.7 (indicative)
evidenced by: Vector ACL Verification Report
GDPR Art. 9
control layer: community mandate — propose objectives
GDPR Art. 88
control layer: community mandate — propose objectives
Take this into your GRC tooling
A control mapping your ISO/IEC 42001 or CSA AICM workbook can ingest, and an Annex IV skeleton to start the technical file from. Indicative mappings only — cells we are not confident about are exported empty rather than filled in.
Standards & Evidence
C2PA Content Credentials
Open technical standard for cryptographically signed content provenance: manifests binding origin, toolchain and edit history to media assets. The de-facto machine-readable implementation path for Art. 50 synthetic-content marking (machine-readable format + detectability duty) — visible labels satisfy the human side, C2PA manifests the machine side. Verification at publication gates produces the disclosure evidence stream.
evidence for: Art. 50
ISO/IEC 42005 (AI Impact Assessment)
Guidance for AI system impact assessments — supports DPIA/FRIA-style analyses.
unverified · no verification date publisher ISO
evidence for: GDPR Art. 35
ISO/IEC 27001:2022 + A.8.28
Information-security management; control A.8.28 (secure coding) is the natural anchor for AI code-generation and QA workflows alongside ISO 42001.
unverified · no verification date publisher ISO
evidence for: GDPR Art. 32
JTC 21 Technical Package (prEN 18228/18229/18281–83)
CEN-CENELEC JTC 21 technical package under standardisation request M/593 (prEN 18228 trustworthiness, 18229 risk management, 18281–83 CV/NLP evaluation et al.); staged drafts, none OJEU-cited yet — Annex III applicability (Dec 2027) is Omnibus-coupled to their availability.
evidence for: EU AI Act
IEEE CertifAIEd™
Ethics certification (transparency, accountability, algorithmic bias, privacy) for products and professionals; interfaces with the EU ALTAI assessment list.
unverified · no verification date
evidence for: EU AI Act
prEN 18229-1 (Trustworthiness Framework, part 1)
Part 1 of the JTC 21 trustworthiness deliverable — the framework layer other prEN 18xxx documents build on.
evidence for: EU AI Act
Evidence you will need (12)
The concrete deliverables this use case's obligations ask for — grouped by what kind of artifact they are. Documentation is the largest single conformity cost block, so the list is a work plan, not a reading list. Full evidence matrix →
Documents & files (2)
Written deliverables an authority or auditor can request as a file.
Post-Market Monitoring Plan & Incident Reportstext-derivedserves 4 obligations
Art. 72 monitoring plan plus Art. 73 serious-incident reports (15 days; 2 days for widespread infringement) — reconciled in one runbook with GDPR Art. 33 (72h), NIS2 (24h/72h) and DORA timelines.
verifiability: documented artefact — verifiable on inspection
chain: Art. 72/73 — Post-Market Monitoring & Incidents · Art. 72/73 — Post-Market Monitoring & Incidents → CO: Performance Monitoring & Drift Management · DORA · NIS2 Directive · Clinical Imaging Triage & Patient Follow-Up · Cross-Border Statutory Tax & Wealth Filing · +2 more
Instructions for Use / Transparency Docstext-derivedserves 2 obligations
Art. 13 deployer-facing documentation: intended purpose, capabilities, limitations, expected accuracy, oversight measures — plus Art. 50 user-facing disclosures.
verifiability: documented artefact — verifiable on inspection
chain: Art. 13 — Transparency to Deployers · Art. 50 — Transparency Duties · Generative Asset Production & Virtual Try-On · Omnichannel Virtual Support & Voice Bots
Assessments (2)
A structured judgement about risk, rights or a management system.
FRIA / AI Impact Assessment (AIIA)text-derivedserves 3 obligations
Fundamental-rights impact assessment (Art. 27, deployer-side) generalized to the AI Impact Assessment: societal, legal and operational risk evaluation per ISO/IEC 42005 and ISO 42001 Clause 8.2, defining HITL intervention parameters and acceptable-use bounds. Cadence: pre-deployment, refreshed annually and on major model updates — a stale AIIA is a finding, not a document.
verifiability: documented artefact — verifiable on inspection
chain: Art. 26 — Deployer Obligations · Art. 27 — Fundamental Rights Impact Assessment · EU AI Act · Clinical Imaging Triage & Patient Follow-Up
Data Protection Impact Assessment (DPIA)text-derivedserves 2 obligations
GDPR Art. 35 assessment for high-risk processing; supervisory-authority consultation where residual risk stays high. ISO/IEC 42005 provides the AI-specific method.
verifiability: documented artefact — verifiable on inspection
chain: § 26 Abs. 1 S. 1 — Erforderlichkeit für Begründung, Durchführung, Beendigung · GDPR Art. 35 — DPIA
Test reports (1)
Measured results from testing, evaluation or red-teaming.
Vector ACL Verification Reportpractice-derived — dispute welcome
Practice-derived artifact: the measured result of probing the retrieval path with low-privilege principals, the ACL reconciliation between source repositories and the index, and the outcome of the response-grounding rights re-check. Records which corpora were probed, which principals were used and every segment that was returned without an entitlement.
verifiability: self-asserted
chain: GDPR Art. 32 — Security of Processing → CO: Vector & Chunk-Level Access Control
Log records (2)
Machine-generated records produced while the system runs.
Event Logs & Decision Tracestext-derivedserves 17 obligations
The single highest-leverage artifact: hash-chained, WORM-stored logs with structured decision traces. Required capability fields per FprEN ISO/IEC 24970: input/output traces, execution timestamps, acting user/agent identity, referenced sources, human overrides. Audit-packet spec per event: model version, system-prompt/context hash, hyper-parameters (temperature, top-p), output payload, confidence score, active policy-ruleset versions, human override record. Simultaneously serves AI Act Art. 12, GDPR accountability, DORA incident reporting, NIS2 logging, PLD disclosure duties and its rebuttable defect presumption; financial-sector regimes push retention to 7 years (SEC 17a-4-class WORM rules). Credibility bar: anchor hash-chain heads externally (qualified timestamp / eIDAS ledger) so integrity survives an insider with admin rights.
verifiability: externally-anchored
chain: Art. 12 — Record-Keeping / Logging · CRA Art. 14 — Vulnerability & Severe-Incident Reporting · DORA Art. 19 — Major ICT-Incident Reporting · GDPR Art. 33/34 — Personal-Data Breach Notification · HIPAA Breach Notification Rule · NIS2 Art. 23 — Significant-Incident Reporting · +21 more
Guardrail Telemetry & Sanitization Recordspractice-derived — dispute welcomeserves 3 obligations
Control-level evidence for the OWASP mappings: guardrail trigger records, blocked-prompt statistics (LLM01), runtime output-sanitization logs (LLM05), groundedness-check outcomes — the empirical proof that declared controls actually execute.
verifiability: tamper-evident
chain: Art. 15 — Accuracy, Robustness, Cybersecurity · Art. 50 — Transparency Duties → CO: AI Interaction & Content Disclosure · Art. 15 — Accuracy, Robustness, Cybersecurity → CO: Runtime Injection Defense · Dynamic Deal Desk & Quoting Engine · Enterprise Marketing Disclosure Compliance · LLM01 Prompt Injection · +3 more
Process records (5)
Traces that a process actually happened, and who did it.
Human-Oversight Protocol & Intervention Recordspractice-derived — dispute welcomeserves 5 obligations
Art. 14 evidence: documented oversight design (gates, thresholds, veto powers), reviewer qualification, and the record of actual approvals, overrides and escalations — also the GDPR Art. 22 meaningful-human-involvement proof.
verifiability: documented artefact — verifiable on inspection
chain: Art. 11 — automated individual decision-making · Art. 14 — Human Oversight · GDPR Art. 22 — Automated Decisions · Art. 12 — Record-Keeping / Logging → CO: Log Access & Retention Governance · Art. 14 — Human Oversight → CO: Oversight Competence & Authority · Clinical Imaging Triage & Patient Follow-Up
Individual Explanation Letters & Counterfactual Recordspractice-derived — dispute welcomeserves 5 obligations
Practice-derived artifact: the issued adverse-decision explanations together with the attribution run, model version and counterfactual scenario that each letter rested on, so an authority or a court can check that the stated reasons are the reasons the system actually used.
verifiability: self-asserted
chain: Art. 11 — automated individual decision-making · Art. 18 — Obligation to assess the creditworthiness of the consumer · Art. 21 — examination of an application · Art. 86 — Right to explanation of individual decision-making · GDPR Art. 22 — Automated Decisions
Vendor & Model Due-Diligence Recordspractice-derived — dispute welcomeserves 4 obligations
DORA Art. 30 / AI Act deployer evidence: scored vendor assessments (jurisdiction, ZDR, BYOK, C5/AIC4/42001 evidence, tenant isolation), contract register, exit strategies for critical third parties.
verifiability: documented artefact — verifiable on inspection
chain: Art. 26 — Deployer Obligations · Art. 26 — Deployer Obligations → CO: Embedded-AI Vendor Governance · DORA · HIPAA (US Health Privacy) · Procurement Variance & Vendor KPI Monitoring
AI Literacy Training Recordspractice-derived — dispute welcomeserves 2 obligations
Art. 4 evidence: role-based training curricula and completion records for staff dealing with AI systems — the one obligation that applies at every risk level.
verifiability: documented artefact — verifiable on inspection
chain: Art. 4 — AI Literacy · Art. 14 — Human Oversight → CO: Oversight Competence & Authority
Personal-Data Breach Notification Recordtext-derivedserves 2 obligations
The GDPR Art. 33(5) record of every personal-data breach: facts, effects, remedial action, plus the notification sent to the supervisory authority and, where required, the data subjects.
verifiability: tamper-evident
chain: GDPR Art. 33/34 — Personal-Data Breach Notification · HIPAA Breach Notification Rule
Architecture Blueprint
Guarded RAG Pattern
For limited-risk conversational/generative systems: deterministic RAG with input rails (DLP/NER, injection blocking), retrieval rails (relevance, freshness, ACL) and output rails (groundedness check with deterministic fallback), plus synthetic-content labelling.
Human-in-the-Loop Core Pattern
For high-risk decision support over people (HR, credit, benefits): confidence-thresholded escalation queues, WORM audit vault, bias testing per ISO 5259 — the human decision is architecturally enforced. CORRECTED 2026-08-15: this pattern previously recommended a post-hoc explainability API (SHAP/LIME or CoT traces) as the explanation mechanism for consequential scoring of natural persons. That is the wrong default. Post-hoc local explainers are sampling-unstable, vary with perturbation choice and are susceptible to fairwashing, and an adverse-action reason that would not change the outcome if the applicant remediated it is not a defensible reason. For consequential scoring (credit, insurance pricing, tenant screening, employment scoring) use intrinsic interpretability — glass-box additive models with monotonic constraints — and keep post-hoc methods as a supplementary diagnostic only.
Mode 1 — Assistant (HITL)
Agent proposes, human disposes: every consequential action reviewed before execution. Default for first deployments, irreversible or legally sensitive actions.
Required Technical Components (27)
Synthetic-Content Labelling / Watermarking
Synthetic-content labelling & watermarking: visible disclosure plus machine-readable provenance (C2PA Content Credentials) embedded in generated images, audio and video; metadata identifying artificial origin survives common transformations. Discharges Art. 50(2)/(4) for deepfakes and synthetic media; verification telemetry (watermark presence/validity checks at publication gates) is the corresponding evidence stream.
from: Art. 50 · Guarded RAG Pattern
Interface Transparency & Content-Marking Layer
The disclosure surface at the engagement layer: an AI-interaction notice on every channel a natural person can reach (web, app, voice, chat, social, marketplace), machine-readable provenance marking on generated or manipulated output, and a disclosure record per interaction that can be produced on request. Sits at the interface, not in the model — a model-side label that the frontend drops is not a disclosure.
from: Art. 50
HITL Escalation Queue & Review UI
HITL escalation queue & review UI ('Human-as-a-Tool': the agent calls the human like any other tool via propose-action objects). Confidence- and risk-threshold routing, SLA timers, structured accept/modify/reject verdicts with digital reviewer signature at gate release — each verdict is itself Art. 14 evidence and feeds the active-learning loop.
from: GDPR Art. 22 · Human-in-the-Loop Core Pattern
Adverse-Decision Reason Generator
Practice-derived component: converts feature attributions (SHAP or an equivalent attribution method) into an individually understandable, legally defensible explanation of an adverse decision — the role the AI system played, the main elements the decision rested on, and counterfactual scenarios stating what would have had to differ for a different outcome. Reason codes are generated from the decisioning path, not from a marketing template, and every issued letter is retained with the model version and the attribution run behind it. Honesty condition: a reason is only usable if acting on it would actually change the outcome, which non-monotonic feature interactions can break (see the post-hoc instability threat).
from: GDPR Art. 22
Bitemporal Memory (GDPR×Art.12)
valid_from/valid_to + transaction time on every record: GDPR erasure removes data from the active retrieval path while the HMAC-chained immutable log survives for Art. 12 / PLD defence; tenant-scoped partitions allow physical scrub of PII.
from: GDPR Art. 17
PII Scrubbing / DLP-NER Layer
Automated detection, pseudonymisation and blocking of personal data in inputs, retrievals and outputs.
from: GDPR Art. 25 · Guarded RAG Pattern
Per-Tenant Retrieval Segmentation
Retrieval is scoped by tenant and by caller entitlement at query time, preventing cross-client and cross-role leakage through shared indexes.
from: GDPR Art. 25
Segmented Vector Store (RBAC + CMEK)
Vector indexes, embeddings and document stores are logically and physically partitioned per client, with role-based access and customer-managed encryption keys.
from: GDPR Art. 25 · GDPR Art. 32
PII/PHI Redaction & Tokenisation Engine
The engine behind inline tokenisation: pre-model interception that replaces identifiers with reversible tokens before a payload leaves the isolation boundary, plus a detokenisation gate that re-identifies only for authorised callers inside the boundary and logs every re-identification. Complements the DLP/NER scrubbing layer, which blocks or masks rather than preserving reversible reference.
from: GDPR Art. 25
DICOM De-Identification Pipeline
Practice-derived component: removal and replacement of identifying attributes in imaging studies before they leave the clinical system — header attributes per the DICOM confidentiality profiles, burned-in pixel text detected and masked, private tags dropped rather than trusted, and a consistent pseudonym per patient so longitudinal studies stay linkable without re-identifying anyone. Re-identification risk on the de-identified corpus is measured, not assumed.
from: GDPR Art. 25 · GDPR Art. 9
Live Risk Register / Posture Management
Continuously updated risk register wired to runtime posture: threat-model deltas, open defects, control status, exposure per system. Includes Shadow-AI discovery — continuous scanning for unsanctioned agents, MCP servers and AI API usage outside the register; an unregistered agent is an unmanaged Art. 12/26 liability and the empirical driver of proportionate (not blanket) controls.
from: GDPR Art. 35
Unified Incident-Response Runbook
One procedure reconciling AI Act Art. 73, GDPR Art. 33 (72h), DORA and NIS2 (24h/72h) timelines and recipients.
from: GDPR Art. 33/34
Retrieval Rails (ACL-aware RAG)
Relevance, freshness and per-user permission checks on every retrieved chunk; curated, versioned index.
from: GDPR Art. 32 · Guarded RAG Pattern
WORM / Immutable Audit Vault
Append-only, hash-chained audit vault (WORM object-lock storage, AES-256 at rest, TLS 1.3 in transit). Guarantees tamper-evidence within the organization's trust domain — which stops your own team, but not an admin who can rebuild the vault. Pair with an external trust anchor and key ceremonies outside the operating team for evidence that holds against the insider scenario.
from: SEC Rule 17a-4 (US Records Retention) · FINRA Rule 4511 (General Books & Records) · Human-in-the-Loop Core Pattern
Agent Memory Record Store
Agent conversational and working memory managed as a books-and-records object rather than a cache: retention schedules per regime, immutability where records rules demand it, deletion paths for erasure requests that do not break the audit trail, and export in a form a supervisor can read. Where an agent's memory carries a business communication or a decision rationale, it is a record — the storage tier does not decide that.
from: SEC Rule 17a-4 (US Records Retention) · FINRA Rule 4511 (General Books & Records)
Multi-Region Failover & Resilience Testing
DORA-grade continuity: regional redundancy, chaos testing, exit strategies for critical third parties.
from: DORA
Vendor & Model Due-Diligence Kit
Scoring model: jurisdiction (CLOUD Act exposure), zero-data-retention, BYOK support, audit evidence (C5/AIC4/ISO 42001/EN 18286:2026), tenant isolation.
from: DORA
Input Rails / Prompt Shields
Pre-model validation of user input: injection detection, topic blocking, encoding checks.
from: Guarded RAG Pattern
Output Rails / Groundedness Check
Faithfulness scoring of answers against retrieved sources; deterministic fallback instead of hallucination; schema-validated structured output.
from: Guarded RAG Pattern
Dual-Gate Validation Pipeline
Input and output validation as two independent gates (MLCommons-hazard-class semantic filters, groundedness checks, structural validators: LLM Guard sub-ms–10 ms, Llama Guard <90 ms, NeMo <50 ms, Guardrails AI 50–200 ms). Latency economics decide the architecture: sequential gate chains add 300–800 ms per agent action; parallel evaluation collapses total added latency to the slowest single check — run independent checks concurrently, reserve sequential ordering for true dependencies.
from: Guarded RAG Pattern
Explainability API (SHAP/LIME/CoT)
Feature attributions for classical ML, reasoning-trace summaries for GenAI — feeds the human reviewer and the technical file.
from: Human-in-the-Loop Core Pattern
Bias Testing & Data Quality Pipeline
Representativeness checks, bias metrics and mitigation per ISO/IEC 5259; versioned datasets with lineage.
from: Human-in-the-Loop Core Pattern
Durable Checkpointing (Pause & Resume)
At oversight gates the complete operational state — working memory, conversation history, tool arguments, intermediate artifacts — is serialized into a durable checkpoint (fast KV store for sub-ms lookups, transactional backend as recovery anchor, vector store for semantic caching of past human decisions). On approval the agent deserializes and resumes at the exact step; matched precedents can shortcut re-planning entirely.
from: Human-in-the-Loop Core Pattern
Kill Switch / Graceful Degradation
Operator stop controls and degraded-mode fallbacks; real-time override (veto) channels for HOTL operation.
from: Human-in-the-Loop Core Pattern
Trust & Risk Dual Scoring
Escalation triggers built from two independent signals, because raw model confidence is uncalibrated: calibrated trust scores (prompt relevance, similarity to historic successes, cross-model consistency) plus deterministic risk scores (sensitive categories, transaction value, protected data) — either crossing its threshold forces human review.
from: Human-in-the-Loop Core Pattern
Active-Learning Feedback Loop
Human corrections at oversight gates are serialized as structured data — original context, model proposal, human edit, rationale — and fed into fine-tuning pipelines and prompt registries, systematically reducing future escalation rates instead of dying in review UIs.
from: Human-in-the-Loop Core Pattern
Confidence Scoring & Threshold Gate
Computes a probabilistic confidence score for every output and holds the transaction when the score falls below the workflow's regulatory threshold.
from: Human-in-the-Loop Core Pattern
Delivery Stack & Pipeline Stage (8)
Service-as-a-Software delivery: the engines, patterns and artifacts this workflow needs on top of the generic obligations. See the full pipeline
Document Intelligence Engine
OCR, layout parsing and semantic clause extraction over filings, contracts and invoices, emitting structured records with span-level source references.
Confidence Scoring & Threshold Gate
Computes a probabilistic confidence score for every output and holds the transaction when the score falls below the workflow's regulatory threshold.
pipeline stage 4 — Output audit & human-in-the-loop gateway
Segmented Vector Store (RBAC + CMEK)
Vector indexes, embeddings and document stores are logically and physically partitioned per client, with role-based access and customer-managed encryption keys.
pipeline stage 1 — Ingestion & data isolation
WORM / Immutable Audit Vault
Append-only, hash-chained audit vault (WORM object-lock storage, AES-256 at rest, TLS 1.3 in transit). Guarantees tamper-evidence within the organization's trust domain — which stops your own team, but not an admin who can rebuild the vault. Pair with an external trust anchor and key ceremonies outside the operating team for evidence that holds against the insider scenario.
pipeline stage 4 — Output audit & human-in-the-loop gateway
Supervisor Attribution Chain
Every model inference, data interaction and client-facing artefact is bound to an authorised supervising natural person — never to a shared service account. Required for SEC Rule 204-2 attribution, SOX segregation of duties and AI Act Art. 26 deployer oversight records.
Confidence-Threshold HITL Routing
Every output carries a confidence score C_s. C_s ≥ θ commits to the immutable ledger and downstream systems; C_s < θ pauses the transaction and routes the payload to a specialist review queue, whose verdict is logged as part of the decision record.
Dual-Gate Validation Pipeline
Input and output validation as two independent gates (MLCommons-hazard-class semantic filters, groundedness checks, structural validators: LLM Guard sub-ms–10 ms, Llama Guard <90 ms, NeMo <50 ms, Guardrails AI 50–200 ms). Latency economics decide the architecture: sequential gate chains add 300–800 ms per agent action; parallel evaluation collapses total added latency to the slowest single check — run independent checks concurrently, reserve sequential ordering for true dependencies.
Per-Tenant Retrieval Segmentation
Retrieval is scoped by tenant and by caller entitlement at query time, preventing cross-client and cross-role leakage through shared indexes.
Build or Buy — Vendor Layer (12)
The graph models vendor CATEGORIES as first-class nodes and keeps named vendors as community-maintained, disputable desc content with lastVerified dates. A category is stable; a vendor list is a currency-layer object like any standard node.
Regulated Foundation-Model Platforms
Frontier commercial APIs and open-weight models under enterprise controls: zero-data-retention tiers, data isolation, fine-tuning governance, safety alignment documentation, EU-sovereign options. Named products live in marketExamples, where the deployment model is recorded in the hosting field rather than asserted in prose. What the class buys you: a model supply relationship with contractual data handling and documentation you can pass to a customer. GPAI-chapter duties and provider due diligence attach at this layer. Selection metrics: see meta.marketLandscape.selectionMetrics.models.
unverified · verified 2026-08-18 community-maintained
selection metrics: ZDR enterprise tiers, data isolation, EU-sovereign options, fine-tuning controls, safety alignment documentation
supplies: Synthetic-Content Labelling / Watermarking
Filters to self-hostable, customer-VPC and open-source options when personal or confidential data cannot leave the EU.
| Example | Sub-category | What it does | Hosting | Claimed alignments |
|---|---|---|---|---|
| OpenAI (Enterprise / API) | proprietary frontier | Enterprise tiers offer zero-data-retention and no-training commitments over the commercial API. Typical: general copilots, document reasoning. | not checked | SOC 2 (claimed)ISO 27001 (claimed)zero-data-retention tier (claimed)GDPR-positioned |
| Anthropic Claude (Enterprise) | proprietary frontier | Enterprise/ZDR tiers with published safety and model documentation practice. Typical: regulated assistants, long-context analysis. | not checked | SOC 2 (claimed)ISO 27001 (claimed)zero-data-retention tier (claimed)HIPAA-eligible (claimed) |
| Google Gemini Enterprise | proprietary frontier | Vertex-hosted frontier models with regional grounding and customer-managed keys. Typical: enterprise search, multimodal workflows. | not checked | SOC 2 (claimed)ISO 27001 (claimed)HIPAA-eligible (claimed)EU data-boundary positioning |
| Cohere | proprietary frontier | Private-cloud and on-prem deployment of retrieval-oriented models. Typical: private RAG, enterprise search. | self-hostable | SOC 2 (claimed) |
and 7 more in the stack advisor →
Community-maintained, disputable examples — not an endorsement and not a ranking. Alignments are as claimed by vendors or the source compilation, not verified by RAIN; a certification is shown as a certification only where a certificate or registry reference is recorded.
Disclosure: RAI·N·avigator operates in this category too, so we have a commercial interest in any comparison here. That is why this layer maps product classes to control objectives and lists named products as community-maintained examples — we publish no rankings, no quadrants and no coverage assertions about any vendor, including ourselves.
Runtime Security & Guardrail Vendors
First-line inline enforcement: single-pass parallel input/output evaluation proxies, injection and exfiltration defense, PII masking, grounding checks, SecOps routing. Named products live in marketExamples; prose here describes the class. What the class buys you: a policy decision point in the request path that fails closed and emits telemetry an auditor can read. Selection metrics: single-pass latency (<20 ms class), catch rates, policy-version telemetry into the AI-BOM. Consolidation matters commercially: a guardrail acquired by a platform vendor tends to follow that platform's roadmap, which is a lock-in question rather than a security one — reported acquisitions are recorded per entry as reported, not asserted here.
unverified · verified 2026-08-18 community-maintained
selection metrics: single-pass parallel evaluation latency (<20 ms class), injection/hallucination catch rates, SecOps/SIEM routing, policy versioning surfaced into the AI-BOM
supplies: Interface Transparency & Content-Marking Layer · Input Rails / Prompt Shields · Output Rails / Groundedness Check · Dual-Gate Validation Pipeline
Filters to self-hostable, customer-VPC and open-source options when personal or confidential data cannot leave the EU.
| Example | Sub-category | What it does | Hosting | Claimed alignments |
|---|---|---|---|---|
| Lakera | guardrail proxy | Inline prompt-injection and content detection at request time. Typical: injection defence, content filtering. | not checked | SOC 2 (claimed)supports Art. 15 robustness measures (claimed) |
| HiddenLayer | model/agent detection & response | Model-layer detection and response with adversarial-attack telemetry. Typical: model threat detection, red-team telemetry. | not checked | SOC 2 (claimed)supports Art. 15 robustness measures (claimed) |
| Palo Alto Prisma AIRS | network-integrated AI security | AI runtime security folded into an existing enterprise network security estate. Typical: enterprise rollout, egress control. | not checked | SOC 2 (claimed)enterprise security integration (claimed) |
| Cisco AI Defense | network-integrated AI security | Discovery of AI usage plus inline enforcement across the corporate network. Typical: shadow-AI discovery, inline enforcement. | not checked | enterprise security integration (claimed) |
and 4 more in the stack advisor →
Community-maintained, disputable examples — not an endorsement and not a ranking. Alignments are as claimed by vendors or the source compilation, not verified by RAIN; a certification is shown as a certification only where a certificate or registry reference is recorded.
Disclosure: RAI·N·avigator operates in this category too, so we have a commercial interest in any comparison here. That is why this layer maps product classes to control objectives and lists named products as community-maintained examples — we publish no rankings, no quadrants and no coverage assertions about any vendor, including ourselves.
Public Transparency Registers & System Cards
Authoring and publishing the outward-facing record: public AI registers, system and model cards, conformity declarations and plain-language notices, with versioning so a published statement can be tied to the system version it described. The register content is produced elsewhere; this class is the publication and version-control surface for it. Selection metrics: see meta.marketLandscape.selectionMetrics.transparency.
unverified · verified 2026-08-17 community-maintained
selection metrics: Versioning of published statements against the system version they describe; whether a card is generated from your governance record or re-authored by hand; language coverage and accessibility of the published surface; export and self-hosting of the public register; whether unpublishing leaves an auditable trail.
supplies: Interface Transparency & Content-Marking Layer
Filters to self-hostable, customer-VPC and open-source options when personal or confidential data cannot leave the EU.
| Example | Sub-category | What it does | Hosting | Claimed alignments |
|---|---|---|---|---|
| Saidot | public AI register | AI register with published system cards and regulation-mapped documentation workflows. Typical: public AI register, system cards. Scope overlap: Its documentation and register scope overlaps this platform's own; we have a commercial interest in the comparison. | SaaS (vendor cloud) | EU AI Act documentation positioningISO 42001 alignment (claimed) |
Community-maintained, disputable examples — not an endorsement and not a ranking. Alignments are as claimed by vendors or the source compilation, not verified by RAIN; a certification is shown as a certification only where a certificate or registry reference is recorded.
Disclosure: RAI·N·avigator operates in this category too, so we have a commercial interest in any comparison here. That is why this layer maps product classes to control objectives and lists named products as community-maintained examples — we publish no rankings, no quadrants and no coverage assertions about any vendor, including ourselves.
Agent Orchestration & SDLC Toolkits
Developer middleware for multi-agent networks, tool-use chains, RAG abstraction, state and memory persistence, and model routing. Named products live in marketExamples; prose here describes the class. Regulatory posture: orchestration code is where autonomy tiering, propose-action objects and fallback routing get implemented — the framework choice constrains which controls are cheap and which are retrofits. Selection metrics: see meta.marketLandscape.selectionMetrics.orchestration.
unverified · verified 2026-08-18 community-maintained
selection metrics: broad model-API abstraction, state/memory management, error recovery, fallback routing hooks
supplies: HITL Escalation Queue & Review UI
Filters to self-hostable, customer-VPC and open-source options when personal or confidential data cannot leave the EU.
| Example | Sub-category | What it does | Hosting | Claimed alignments |
|---|---|---|---|---|
| LangChain / LangGraph | agent framework | Graph-structured agent runtime; interrupt/pause nodes support implementing human approval at defined steps. Typical: multi-step agents, approval workflows. | not checked | supports implementing Art. 14 oversight (claimed)supports Art. 12 step logging (claimed) |
| LlamaIndex | RAG framework | Indexing and query abstractions over documents and structured sources. Typical: enterprise RAG, document agents. | open source | retrieval-governance positioning |
| Microsoft AutoGen | multi-agent framework | Conversational multi-agent patterns with pluggable tool executors. Typical: multi-agent research, code agents. | not checked | research/OSS, no vendor certification |
| CrewAI | multi-agent framework | Role-based agent teams with task delegation and process templates. Typical: process automation, role-based agents. | not checked | vendor-stated security posture |
and 6 more in the stack advisor →
Community-maintained, disputable examples — not an endorsement and not a ranking. Alignments are as claimed by vendors or the source compilation, not verified by RAIN; a certification is shown as a certification only where a certificate or registry reference is recorded.
Disclosure: RAI·N·avigator operates in this category too, so we have a commercial interest in any comparison here. That is why this layer maps product classes to control objectives and lists named products as community-maintained examples — we publish no rankings, no quadrants and no coverage assertions about any vendor, including ourselves.
Agent Observability & Model Risk Management
Tracing, evaluation, drift monitoring and model-validation records. This layer is where Art. 12 record-keeping becomes technically real (step-level traces, prompt/response records, retention control) and where model-risk practice in the SR 11-7 tradition — validation evidence, performance and drift monitoring, challenger comparison — is operated. Gateways and tracing tools produce the logs; the retention, integrity and access regime around them is still yours.
unverified · verified 2026-08-18 community-maintained
selection metrics: Trace completeness per agent step; log retention and immutability options; drift/quality metrics available out of the box; evaluation dataset support; export into your audit vault; self-host option.
supplies: Adverse-Decision Reason Generator · Dual-Gate Validation Pipeline · Explainability API (SHAP/LIME/CoT) · Bias Testing & Data Quality Pipeline · Confidence Scoring & Threshold Gate
Filters to self-hostable, customer-VPC and open-source options when personal or confidential data cannot leave the EU.
| Example | Sub-category | What it does | Hosting | Claimed alignments |
|---|---|---|---|---|
| LangSmith | agent tracing & evaluation | Trace capture and evaluation over LangChain/LangGraph runs with dataset-based scoring. Typical: step tracing, regression evaluation. | not checked | SOC 2 (claimed)supports Art. 12 record-keeping (claimed) |
| Langfuse | agent tracing & evaluation | Open-source tracing, prompt management and evaluation; self-hostable for retention control. Typical: self-hosted tracing, cost/latency analytics. | open source | GDPR-positionedsupports Art. 12 record-keeping (claimed) |
| Arize AI / Phoenix | ML & LLM observability | Production monitoring with drift and performance analysis; Phoenix is the open-source tracing side. Typical: drift monitoring, production analytics. | not checked | SOC 2 (claimed)drift-monitoring positioning (SR 11-7 style, claimed) |
| Helicone | LLM gateway & logging | Proxy-level logging of prompts, costs and latency across providers. Typical: gateway logging, cost control. | not checked | SOC 2 (claimed)supports Art. 12 record-keeping (claimed) |
and 11 more in the stack advisor →
Community-maintained, disputable examples — not an endorsement and not a ranking. Alignments are as claimed by vendors or the source compilation, not verified by RAIN; a certification is shown as a certification only where a certificate or registry reference is recorded.
Disclosure: RAI·N·avigator operates in this category too, so we have a commercial interest in any comparison here. That is why this layer maps product classes to control objectives and lists named products as community-maintained examples — we publish no rankings, no quadrants and no coverage assertions about any vendor, including ourselves.
AI GRC & Governance Platforms
Second-line systems of record: model/agent inventory incl. third-party SaaS AI, automated risk tiering, policy administration, cross-framework mapping and control deduplication, audit-evidence generation, intake workflows. Named products live in marketExamples, which is the single source of truth for this layer — prose here describes the class, not the field. What the class buys you: one register a second line can defend, and evidence assembled once and reused across frameworks. Selection metrics: see meta.marketLandscape.selectionMetrics.grc. One compilation-reported item is deliberately kept as unverified: a claimed updated US banking model-risk guidance 'SR 26-2'. Two secondary compilations repeating it is corroboration of the rumour, not of the guidance; it stays flagged pending verification against Federal Reserve primary sources, and a curator verification proposal is filed. All alignments in this layer are vendor-positioned claims, never certifications.
unverified · verified 2026-08-18 community-maintained
selection metrics: multi-model/multi-cloud cataloging incl. third-party SaaS, automated risk tiering, regulatory reporting, independent-2nd-line deployability, cross-framework control deduplication
supplies: Adverse-Decision Reason Generator · Live Risk Register / Posture Management · Vendor & Model Due-Diligence Kit
Filters to self-hostable, customer-VPC and open-source options when personal or confidential data cannot leave the EU.
| Example | Sub-category | What it does | Hosting | Claimed alignments |
|---|---|---|---|---|
| Credo AI | AI governance platform | Policy packs, risk tiering and evidence workflows mapped across frameworks. Typical: AI registry, policy administration. Scope overlap: Its scope overlaps this platform's own; we have a commercial interest in the comparison. | not checked | ISO 42001 alignment (claimed)EU AI Act readiness positioning |
| Holistic AI | AI governance & audit | Risk assessment, bias auditing and regulatory reporting workflows. Typical: bias audit, regulatory reporting. Scope overlap: Its scope overlaps this platform's own; we have a commercial interest in the comparison. | not checked | NYC LL144 audit support (claimed)EU AI Act readiness positioning |
| IBM watsonx.governance | AI governance platform | Governance, factsheets and monitoring integrated with the IBM stack. Typical: factsheets, model monitoring. Scope overlap: Its scope overlaps this platform's own; we have a commercial interest in the comparison. | not checked | ISO 42001 alignment (claimed)Art. 11 documentation support (claimed) |
| ModelOp | AI/model governance | Model and agent inventory with automated lifecycle controls for large estates. Typical: model inventory, control automation. Scope overlap: Its scope overlaps this platform's own; we have a commercial interest in the comparison. | not checked | model-risk positioning (SR 11-7 style, claimed)ISO 42001 alignment (claimed) |
and 3 more in the stack advisor →
Community-maintained, disputable examples — not an endorsement and not a ranking. Alignments are as claimed by vendors or the source compilation, not verified by RAIN; a certification is shown as a certification only where a certificate or registry reference is recorded.
Disclosure: RAI·N·avigator operates in this category too, so we have a commercial interest in any comparison here. That is why this layer maps product classes to control objectives and lists named products as community-maintained examples — we publish no rankings, no quadrants and no coverage assertions about any vendor, including ourselves.
Grounding, Retrieval & Agent Memory
The grounding layer between raw sources and the model: document parsers, embedding models, vector databases and — new in the agentic era — persistent agent memory stores. Memory is the hard part: once a personal fact is embedded, GDPR Art. 17 erasure has to reach the vector and the memory record, not just the source row, and embeddings are partially reconstructable (see IronCore in the privacy layer). Retrieval quality is also a data-governance question under Art. 10: what got parsed, chunked and indexed is what the system 'knows'.
unverified · verified 2026-08-18 community-maintained
selection metrics: Parsing fidelity on your worst document class; retrieval precision/recall on a labelled set; tenant and ACL isolation model; per-vector encryption and erasure path; memory TTL and record semantics; self-host option.
supplies: Bitemporal Memory (GDPR×Art.12) · Per-Tenant Retrieval Segmentation · Segmented Vector Store (RBAC + CMEK) · Retrieval Rails (ACL-aware RAG) · Agent Memory Record Store
Filters to self-hostable, customer-VPC and open-source options when personal or confidential data cannot leave the EU.
| Example | Sub-category | What it does | Hosting | Claimed alignments |
|---|---|---|---|---|
| Docling | document parser | Open-source layout-aware parsing of PDFs and office formats into structured chunks. Typical: RAG ingestion, air-gapped pipelines. | self-hostable | EU sovereignty positioning |
| LlamaParse | document parser | Managed parsing service tuned for tables and complex documents feeding RAG. Typical: RAG ingestion, table extraction. | not checked | SOC 2 (claimed) |
| Amazon Textract | document parser | OCR and form/table extraction with per-page pricing inside AWS. Typical: document intake, claims processing. | not checked | SOC 2 (claimed)HIPAA-eligible (claimed)ISO 27001 (claimed) |
| Diffbot | web/knowledge extraction | Structured extraction and knowledge-graph construction from web sources. Typical: market monitoring, entity resolution. | not checked | vendor-stated security posture |
and 12 more in the stack advisor →
Community-maintained, disputable examples — not an endorsement and not a ranking. Alignments are as claimed by vendors or the source compilation, not verified by RAIN; a certification is shown as a certification only where a certificate or registry reference is recorded.
Disclosure: RAI·N·avigator operates in this category too, so we have a commercial interest in any comparison here. That is why this layer maps product classes to control objectives and lists named products as community-maintained examples — we publish no rankings, no quadrants and no coverage assertions about any vendor, including ourselves.
Confidential Computing & Privacy Engines
Data-in-use protection and pre-model privacy interception: enclave and runtime encryption, key management, tokenisation vaults, PII detection and redaction, application-layer and vector encryption. Named products live in marketExamples; prose here describes the class. Select on: enclave attestation support, key custody model (external HSM / BYOK), detokenisation audit trail, latency added per call, and coverage of the identifier classes your regime actually names. Selection metrics: see meta.marketLandscape.selectionMetrics.privacy.
unverified · verified 2026-08-18 community-maintained
selection metrics: enclave attestation support, key custody (external HSM / BYOK), detokenisation audit trail, added latency per call, coverage of the identifier classes your regime names, in-boundary deployment option
supplies: PII/PHI Redaction & Tokenisation Engine · DICOM De-Identification Pipeline
Filters to self-hostable, customer-VPC and open-source options when personal or confidential data cannot leave the EU.
| Example | Sub-category | What it does | Hosting | Claimed alignments |
|---|---|---|---|---|
| Anjuna | confidential computing | Runs workloads inside hardware enclaves without application rewrites. Typical: data-in-use protection, regulated inference. | not checked | confidential-computing positioningDORA-positioned (claimed) |
| Fortanix | confidential computing & KMS | Enclave runtime plus key management and tokenisation services. Typical: key management, data-in-use protection. | not checked | FIPS 140-2 (claimed)DORA-positioned (claimed)HIPAA-positioned (claimed) |
| Skyflow | privacy vault | Polymorphic data vault de-identifying records before they reach a model. Typical: PII vaulting, pre-model redaction. | not checked | SOC 2 (claimed)HIPAA-positionedGDPR-positioned |
| Private AI | PII detection & redaction | Detection and redaction of identifiers across text, documents and audio. Typical: inline redaction, document de-identification. | not checked | GDPR-positionedHIPAA-positioned |
and 1 more in the stack advisor →
Community-maintained, disputable examples — not an endorsement and not a ranking. Alignments are as claimed by vendors or the source compilation, not verified by RAIN; a certification is shown as a certification only where a certificate or registry reference is recorded.
Disclosure: RAI·N·avigator operates in this category too, so we have a commercial interest in any comparison here. That is why this layer maps product classes to control objectives and lists named products as community-maintained examples — we publish no rankings, no quadrants and no coverage assertions about any vendor, including ourselves.
Secure Data Infrastructure & Vector Storage
Governed retrieval substrate: vector databases, lakehouses and catalogs with tenant/namespace isolation, RBAC and client-managed keys (CMEK), lineage into RAG chunks, air-gap options, and code-level data and AI lineage. Named products live in marketExamples; prose here describes the class. What the class buys you: retrieval that can be scoped per requester and traced back to a source record. The Art. 10 runtime data-governance duties land here. Selection metrics: see meta.marketLandscape.selectionMetrics.data.
unverified · verified 2026-08-18 community-maintained
selection metrics: namespace/tenant isolation, RBAC + CMEK, lineage into RAG chunks, SOC 2 / ISO 27001 attestations, air-gap capability
supplies: Retrieval Rails (ACL-aware RAG) · Agent Memory Record Store
Filters to self-hostable, customer-VPC and open-source options when personal or confidential data cannot leave the EU.
| Example | Sub-category | What it does | Hosting | Claimed alignments |
|---|---|---|---|---|
| Azure AI Search | managed retrieval | Managed hybrid search with security trimming against tenant identities. Typical: ACL-aware RAG, enterprise search. | not checked | ISO 27001 (claimed)SOC 2 (claimed) |
| Databricks Unity Catalog | governed lakehouse | Catalog and lineage spanning tables, features and RAG chunks. Typical: lineage evidence, governed RAG. | not checked | SOC 2 (claimed)lineage/Art. 10 support (claimed) |
| Relyance AI | code-level data & AI lineage | Parses source repositories to map data and inference flows at code level, with CI checks on changes to those flows. Typical: data lineage, shift-left privacy review. Scope overlap: Its AI-governance reporting scope overlaps this platform's own; we have a commercial interest in the comparison. | SaaS (vendor cloud) | GDPR programme tooling (claimed)EU AI Act readiness positioning |
| Snowflake Cortex | governed lakehouse | Model calls inside the warehouse boundary with masking and clean rooms. Typical: in-warehouse inference, governed analytics. | not checked | SOC 2 (claimed)ISO 27001 (claimed)HIPAA-eligible (claimed) |
Community-maintained, disputable examples — not an endorsement and not a ranking. Alignments are as claimed by vendors or the source compilation, not verified by RAIN; a certification is shown as a certification only where a certificate or registry reference is recorded.
Disclosure: RAI·N·avigator operates in this category too, so we have a commercial interest in any comparison here. That is why this layer maps product classes to control objectives and lists named products as community-maintained examples — we publish no rankings, no quadrants and no coverage assertions about any vendor, including ourselves.
Cryptographic Evidence & Audit Ledger
Tamper-evident recording of what a system did: content-addressed decision records, hash chains and external anchoring, so a log can be shown not to have been rewritten after the fact. This is the layer that turns Art. 12 logging and Art. 19 retention from a storage question into an evidentiary one. AI Verify is carried in RAIN as a STANDARD node (sg-ai-verify), not duplicated here as a vendor.
unverified · verified 2026-08-18 community-maintained
selection metrics: Append-only guarantees and who can rotate or delete (including the vendor); anchoring mechanism (qualified timestamp, transparency log, notarisation) and whether verification works without the vendor; retention and export in a readable format at end of contract; throughput and cost at your event volume.
supplies: WORM / Immutable Audit Vault
Filters to self-hostable, customer-VPC and open-source options when personal or confidential data cannot leave the EU.
| Example | Sub-category | What it does | Hosting | Claimed alignments |
|---|---|---|---|---|
| Fact0 | cryptographic evidence ledger | Positions itself as a tamper-evident ledger for AI decision records. Typical: decision records, audit trail. | not checked | supports Art. 12 record-keeping (claimed) |
| Traccia | audit trail & traceability | Positions itself around traceability of AI pipeline steps and artefacts. Typical: traceability, artifact lineage. | not checked | supports Art. 12 record-keeping (claimed) |
Community-maintained, disputable examples — not an endorsement and not a ranking. Alignments are as claimed by vendors or the source compilation, not verified by RAIN; a certification is shown as a certification only where a certificate or registry reference is recorded.
Disclosure: RAI·N·avigator operates in this category too, so we have a commercial interest in any comparison here. That is why this layer maps product classes to control objectives and lists named products as community-maintained examples — we publish no rankings, no quadrants and no coverage assertions about any vendor, including ourselves.
Sovereign Infrastructure
Compute and storage under EU jurisdictional control. Two structurally different offers, and the difference is the decision: native EU providers give full jurisdictional isolation with narrower service catalogs and thinner managed-AI tooling; hyperscaler sovereign constructions give the broad catalog with contractual and operational isolation, where the residual question is the control plane, support access and operational metadata rather than the data plane. Named offers live in marketExamples, which is the single source of truth for this layer — prose here describes the class, not the field. Claimed alignments recorded per entry: positioning for BSI C5 / C3A and ANSSI SecNumCloud attestation, NIS2 and DORA third-party requirements. Nothing here is an endorsement, and no provider in this category is 'CLOUD-Act-proof' by label alone — ask who holds the keys and who administers the plane.
unverified · verified 2026-08-18 community-maintained
selection metrics: jurisdiction of the control plane (not only the data plane), operator nationality and support-access paths, key custody, C5 / C3A / SecNumCloud attestation scope, managed-AI service depth vs. isolation trade-off, exit and repatriation terms
supplies: Multi-Region Failover & Resilience Testing
Filters to self-hostable, customer-VPC and open-source options when personal or confidential data cannot leave the EU.
| Example | Sub-category | What it does | Hosting | Claimed alignments |
|---|---|---|---|---|
| OVHcloud | native EU | French provider with EU-only jurisdiction and a narrower managed-AI catalog than the hyperscalers. Typical: EU-resident inference, regulated workload hosting. | not checked | ISO 27001 (claimed)SecNumCloud-positionedGDPR-positioned |
| Scaleway | native EU | EU-operated cloud with GPU instances and managed inference under French corporate control. Typical: EU-resident inference, fine-tuning. | not checked | ISO 27001 (claimed)GDPR-positioned |
| STACKIT | native EU | German provider (Schwarz Group) positioned for data residency in Germany. Typical: public sector, retail data platforms. | not checked | C5-positionedGDPR-positioned |
| AWS European Sovereign Cloud | sovereign hyperscaler | Separately operated EU region set with EU-resident personnel and keys; full hyperscaler catalog. Typical: large-scale enterprise AI, regulated hosting. | not checked | ISO 27001 (claimed)SOC 2 (claimed)EU data-boundary positioning |
and 11 more in the stack advisor →
Community-maintained, disputable examples — not an endorsement and not a ranking. Alignments are as claimed by vendors or the source compilation, not verified by RAIN; a certification is shown as a certification only where a certificate or registry reference is recorded.
Disclosure: RAI·N·avigator operates in this category too, so we have a commercial interest in any comparison here. That is why this layer maps product classes to control objectives and lists named products as community-maintained examples — we publish no rankings, no quadrants and no coverage assertions about any vendor, including ourselves.
Runtime Guardrails & Enforcement
Policy enforcement in the request path: input/output validation, injection and exfiltration defence, structured-output constraints and action blocking. Distinct from observability layers because these products are in-line and can refuse. Selection questions: added latency at p95, whether enforcement is fail-open or fail-closed, whether policies are versioned artefacts, and whether the layer can be self-hosted inside your data boundary.
unverified · verified 2026-08-18 community-maintained
selection metrics: Where enforcement sits (inline proxy, sidecar, SDK) and the added latency at your token volumes; whether policy is versioned and testable as code; fail-open vs. fail-closed behaviour under guardrail outage; language and modality coverage; whether every block writes an evidence record you can cite later.
supplies: Input Rails / Prompt Shields · Output Rails / Groundedness Check
Filters to self-hostable, customer-VPC and open-source options when personal or confidential data cannot leave the EU.
| Example | Sub-category | What it does | Hosting | Claimed alignments |
|---|---|---|---|---|
| Guardrails AI | validation framework | Open-source validator framework for structured output and content policies in the request path. Typical: output validation, structured output. | open source | supports Art. 15 robustness measures (claimed) |
| NVIDIA NeMo Guardrails | dialogue policy rails | Programmable dialogue and topic rails placed around an LLM application. Typical: topic control, dialogue policy. | open source | supports Art. 50 interaction disclosure patterns (claimed) |
| Lakera AI | guardrail proxy | Inline prompt-injection and content detection at request time. Typical: injection defence, content filtering. | SaaS (vendor cloud) | SOC 2 (claimed)supports Art. 15 robustness measures (claimed) |
| Credal AI | enterprise access & policy layer | Permission-aware access layer with data-loss controls in front of enterprise assistants. Typical: access control, DLP. | SaaS (vendor cloud) | SOC 2 (claimed) |
Community-maintained, disputable examples — not an endorsement and not a ranking. Alignments are as claimed by vendors or the source compilation, not verified by RAIN; a certification is shown as a certification only where a certificate or registry reference is recorded.
Disclosure: RAI·N·avigator operates in this category too, so we have a commercial interest in any comparison here. That is why this layer maps product classes to control objectives and lists named products as community-maintained examples — we publish no rankings, no quadrants and no coverage assertions about any vendor, including ourselves.
Procurement rule: Derived from three-lines-of-defense separation: the second-line GRC platform must be procured and deployed independently of any first-line runtime or model vendor — a governance tool that only sees its own vendor's models cannot govern a multi-model estate, and closed third-party SaaS AI can only be governed contractually (intake, attestation, AI-BOM disclosure), never by inline inspection.
Outsourced delivery BPO · SaaS · Service-as-a-Software caveats
Delivery Model — BPO · SaaS · Service-as-a-Software
Spectrum
BPO: input-priced (billable hours/FTEs), linear headcount scaling, human error & attrition as primary risk
SaaS: capability-priced (software access), client operates the workload, implementation/adoption failure as primary risk
Service-as-a-Software: outcome-priced (SLA on completed work), provider-managed AI executes 60–80% of cognitive tasks with specialist supervision, algorithmic bias & non-compliance as primary risk
Caveats in regulated markets
— Outcome SLAs move compliance risk onto the provider — but NOT the buyer's deployer duties: Art. 26 oversight, log retention and FRIA obligations stay with the enterprise even when execution is outsourced.
— Provider role analysis is the central legal question: a productized platform that fine-tunes, re-purposes or chains models can flip into the Art. 25 provider role with full high-risk obligations.
— Certified operations (ISO 42001) function as a procurement moat and shortcut third-party risk assessment — but organizational certificate ≠ product conformity (never conflate, see meta.assuranceEcosystem).
— The buyer's evidence chain must reach into the provider: contractually mandated AI-BOM disclosure, ZDR certificates, bias-audit reports and logging-ledger access are the artifacts that make an outsourced workflow auditable.
Threat Profile
Hallucinated Legal Terms
Extraction models invent clause values, dates or thresholds that look plausible and enter downstream risk systems unchallenged.
mitigate with: Output Rails / Groundedness Check, Confidence Scoring & Threshold Gate
RAG Context Loss & Truncation
Long agreements are chunked such that defining context (definitions, annexes, amendments) is lost, so extracted terms are locally correct but globally wrong.
mitigate with: Retrieval Rails (ACL-aware RAG), Data Lineage & Versioning
LLM09 Misinformation
Hallucinated or wrong outputs create liability and decision risk.
mitigate with: Output Rails / Groundedness Check, Explainability API (SHAP/LIME/CoT)
Permission Loss in Embedding Pipelines
When documents are chunked and embedded, the access-control list of the source system is not carried into the vector store: the chunk keeps its text and loses its permissions. Semantic similarity search then returns segments the requesting user was never entitled to read, so an unauthorised user can reconstruct protected content or third-party personal data out of retrieved fragments without ever touching the source repository. This is a security-of-processing failure under GDPR Art. 32 and a trade-secret exposure, and it applies to every RAG deployment, not only to those handling special-category data.
mitigate with: CO: Vector & Chunk-Level Access Control, Retrieval Rails (ACL-aware RAG), Segmented Vector Store (RBAC + CMEK)