> ## Documentation Index
> Fetch the complete documentation index at: https://docs.sint.gg/llms.txt
> Use this file to discover all available pages before exploring further.

# Compliance

> IEC 62443, EU AI Act, NIST AI RMF, ISO/IEC 42001 mappings.

SINT Protocol is designed to map cleanly to the three compliance frameworks that will matter for high-risk AI deployments in 2026 and beyond. The tier crosswalk is machine-readable at `GET /v1/compliance/tier-crosswalk`.

<Warning>
  **EU AI Act high-risk provisions take effect 2 August 2026.** Article 13 (transparency) and Article 14 (human oversight) are the two most directly addressed by SINT. Penalties reach €35M or 7% of global annual turnover.
</Warning>

## IEC 62443 — FR1 through FR7

IEC 62443 is the canonical industrial-control-systems security series. IEC PAS 62443-1-6:2025 (published December 2025) extends it to Industrial IoT, which is where physical AI lives.

| FR      | Requirement                     | SINT mechanism                                                                                          |
| ------- | ------------------------------- | ------------------------------------------------------------------------------------------------------- |
| **FR1** | Identification & Authentication | `SintCapabilityToken` with Ed25519 identity; W3C DID portability; challenge-response in `PENDING_AUTH`. |
| **FR2** | Use Control                     | Four-tier gate (T0–T3); `scope.allowedTools` allowlists; per-resource action restrictions.              |
| **FR3** | System Integrity                | SHA-256 hash-chained Evidence Ledger; `ProofReceipt` with TEE attestation for T2/T3.                    |
| **FR4** | Data Confidentiality            | TLS 1.3 on all inter-component traffic; scoped tokens prevent sensor access without explicit grant.     |
| **FR5** | Restricted Data Flow            | Policy Gateway allowlists; `geofence` constraint; per-topic SINT Bridge DFA.                            |
| **FR6** | Timely Response to Events       | `safety.estop.triggered` event; E-stop universality (I-G2); T0 p50 ≤ 1ms.                               |
| **FR7** | Resource Availability           | Per-token rate limiting; `maxRepetitions` constraint; budget enforcement in capsule sandbox.            |

## EU AI Act — Article 13 and Article 14

<Note>
  **Verifiable transparency, not claimed transparency.** The industry status quo is application-layer logs that are deletable, editable, and whose integrity rests on operational discipline. SINT's Evidence Ledger moves transparency from *claimed* to *verifiable*: append-only, SHA-256 hash-chained, TEE-signed per event, with deletion detectable by external audit.
</Note>

| Requirement                              | Industry status quo                     | SINT implementation                                                                                 |
| ---------------------------------------- | --------------------------------------- | --------------------------------------------------------------------------------------------------- |
| **Art. 13** log keeping for high-risk AI | Application logs, deletable             | Append-only ledger, SHA-256 chain, deletion detectable                                              |
| **Art. 13** decision traceability        | Human-readable audit trails             | Machine-verifiable `decisionRef` linkage; TEE signature per event                                   |
| **Art. 14** human oversight              | "Human in the loop" as policy statement | Formal T3 state; ESCALATING → HUMAN\_APPROVED/DENIED/TIMEOUT                                        |
| **Art. 14(4)(e)** override mechanism     | Kill switch procedure                   | `CircuitBreaker.trip()` with I-G2 universality; manual reset required                               |
| **Art. 17** quality management           | QMS documentation                       | [157-test conformance suite](/protocol/conformance); `AUDIT_CHAIN_VERIFIED` triggers external audit |
| **Art. 9** risk management               | Risk registers                          | Tier escalation function operationalizes continuous risk assessment                                 |

## NIST AI RMF

NIST's AI Risk Management Framework organizes risk into four functions: GOVERN, MAP, MEASURE, MANAGE.

| Function    | NIST control                          | SINT mechanism                                    |
| ----------- | ------------------------------------- | ------------------------------------------------- |
| **GOVERN**  | GOVERN-1.1 (human oversight policies) | T2/T3 escalation; CircuitBreaker                  |
| **GOVERN**  | GOVERN-1.2 (risk tolerance)           | Per-deployment `SintDeploymentProfile`            |
| **GOVERN**  | GOVERN-4.1 (accountability)           | EvidenceLedger hash-chained append-only           |
| **MAP**     | MAP-1.1 (risk identification)         | PolicyGateway tier assignment                     |
| **MAP**     | MAP-3.5                               | CSML risk scoring                                 |
| **MAP**     | MAP-5.1 (stakeholder impact)          | `PhysicsConstraints` with velocity/force/geofence |
| **MEASURE** | MEASURE-1.1 (risk measurement)        | CSML score                                        |
| **MEASURE** | MEASURE-2.5 (bias/fairness)           | Per-model CSML                                    |
| **MEASURE** | MEASURE-2.6 (monitoring)              | EvidenceLedger with SIEM export                   |
| **MEASURE** | MEASURE-2.8 (operation)               | PolicyGateway benchmark suite                     |
| **MANAGE**  | MANAGE-1.3 (risk response)            | Tier escalation                                   |
| **MANAGE**  | MANAGE-2.2 (treatment)                | Token attenuation                                 |
| **MANAGE**  | MANAGE-4.2 (incident response)        | CircuitBreakerPlugin; forensic trail              |

Full mapping with sub-category evidence: [`docs/specs/nist-ai-rmf-crosswalk.md`](https://github.com/sint-ai/sint-protocol/blob/main/docs/specs/nist-ai-rmf-crosswalk.md).

## ISO/IEC 42001

| SINT tier  | ISO/IEC 42001 clause                         |
| ---------- | -------------------------------------------- |
| T0 Observe | Clause 9 + Clause 8 controls                 |
| T1 Prepare | Clause 8.1 / 8.2 operational risk treatment  |
| T2 Act     | Clause 8 + Clause 6 operational controls     |
| T3 Commit  | Clause 8.3 + Clause 10 corrective governance |

## OWASP Top 10 for Agentic Applications

All ten ASI categories are addressed with dedicated regression tests in the [conformance suite](/protocol/conformance).

| Category                       | SINT control                                                 |
| ------------------------------ | ------------------------------------------------------------ |
| **ASI01** Goal Hijacking       | GoalHijackPlugin (5-layer heuristics) + CSML drift detection |
| **ASI02** Prompt Injection     | T3 classification, token scoping                             |
| **ASI03** Insecure Output      | Transformation constraints                                   |
| **ASI04** Tool Misuse          | PolicyGateway tier assignment                                |
| **ASI05** Shell via Tool Calls | T3 CRITICAL classifier                                       |
| **ASI06** Memory Poisoning     | MemoryIntegrityChecker                                       |
| **ASI07** Excessive Autonomy   | T2 escalation threshold                                      |
| **ASI08** Inadequate Logging   | EvidenceLedger mandatory recording                           |
| **ASI09** Supply Chain         | CapabilityToken chain validation                             |
| **ASI10** Rogue Agent          | CircuitBreakerPlugin (EU AI Act stop button)                 |

## Machine-readable crosswalk

```bash theme={null}
curl https://api.sint.gg/v1/compliance/tier-crosswalk
```

Returns the full mapping as JSON for integration into compliance automation tools.

<Tip>
  **Preparing for a compliance audit?** The [conformance suite](/protocol/conformance) generates a per-invariant certification bundle that auditors can ingest directly. See [`docs/reports/certification-bundle-summary.md`](https://github.com/sint-ai/sint-protocol/blob/main/docs/reports/certification-bundle-summary.md) in the repo.
</Tip>