CSOAI   Home · Journal · Certification · Fabric
The 52-Article Charter · 28 of 52 · full text

Article 28: Interoperability Standards

Published from the canonical CSOAI Partnership Charter (effective 15 January 2026). Full text below.

Version: 1.0 Effective Date: January 15, 2026, 09:00 GMT Status: Technical Article - Integration Standards

Framework Integration: ISO/IEC 42001 AI Management System, CEN-CENELEC Harmonized Standards, NIST AI RMF GOVERN Function


PREAMBLE

This Article establishes comprehensive interoperability standards for AI systems. AI systems do not exist in isolation—they integrate with other systems, share data, and must comply with diverse regulatory frameworks. Interoperability enables ecosystem, compliance, and safety.

Core Principle: Open standards, seamless integration, regulatory harmony.


28.1 API STANDARDS

28.1.1 REST API Requirements

Representational State Transfer:

Mandatory for All AI Systems with External Interfaces:

Design Principles:

URL Structure: ``` https://api.company.com/v{version}/{resource}/{id}/{action}

Examples: GET /v1/models # List models GET /v1/models/{id} # Get specific model POST /v1/models/{id}/predict # Make prediction GET /v1/models/{id}/metrics # Get performance metrics POST /v1/compliance/verify # Verify compliance ```

HTTP Methods:

| Method | Purpose | Idempotent | Safe |
|--------|---------|------------|------|
| GET | Retrieve resource | Yes | Yes |
| POST | Create resource / Action | No | No |
| PUT | Replace resource | Yes | No |
| PATCH | Partial update | No | No |
| DELETE | Remove resource | Yes | No |

Status Codes:

| Code | Meaning | Use Case |
|------|---------|----------|
| 200 | OK | Successful GET, PUT, PATCH |
| 201 | Created | Successful POST (resource created) |
| 204 | No Content | Successful DELETE |
| 400 | Bad Request | Invalid input |
| 401 | Unauthorized | Missing/invalid authentication |
| 403 | Forbidden | Authenticated but not authorized |
| 404 | Not Found | Resource doesn't exist |
| 429 | Too Many Requests | Rate limit exceeded |
| 500 | Internal Server Error | Server-side failure |
| 503 | Service Unavailable | Maintenance, overload |

Response Format (JSON): ```json { "data": { "id": "model_123", "type": "prediction", "attributes": { "label": "cat", "confidence": 0.95, "timestamp": "2026-01-15T10:30:00Z" } }, "meta": { "request_id": "req_abc123", "processing_time_ms": 45 } } ```

Error Response: ```json { "error": { "code": "INVALID_INPUT", "message": "Image format not supported", "details": "Expected JPEG or PNG, received GIF", "request_id": "req_abc123", "documentation": "https://api.company.com/docs/errors#INVALID_INPUT" } } ```

28.1.2 OpenAPI Specification

Machine-Readable API Documentation:

Required for All Public APIs:

OpenAPI 3.1 Compliance: ```yaml openapi: 3.1.0 info: title: AI Model API version: 1.0.0 description: API for AI model inference and management contact: email: api@company.com license: name: Commercial servers: - url: https://api.company.com/v1 description: Production - url: https://api-staging.company.com/v1 description: Staging

paths: /predict: post: summary: Make a prediction operationId: predict tags: - Inference requestBody: required: true content: application/json: schema: $ref: '#/components/schemas/PredictRequest' responses: '200': description: Successful prediction content: application/json: schema: $ref: '#/components/schemas/PredictResponse' '400': $ref: '#/components/responses/BadRequest' '401': $ref: '#/components/responses/Unauthorized' components: schemas: PredictRequest: type: object required: - input properties: input: type: string description: Input data (base64 encoded for binary) options: type: object properties: return_confidence: type: boolean default: true PredictResponse: type: object properties: prediction: type: string confidence: type: number format: float latency_ms: type: integer securitySchemes: ApiKeyAuth: type: apiKey in: header name: X-API-Key security: - ApiKeyAuth: [] ```

Auto-Generated Documentation:

28.1.3 gRPC for High-Performance

When REST Not Sufficient:

Use Cases:

Protocol Buffers Definition: ```protobuf syntax = "proto3";

package ai.inference.v1;

service InferenceService { rpc Predict(PredictRequest) returns (PredictResponse); rpc StreamPredict(stream PredictRequest) returns (stream PredictResponse); rpc BatchPredict(BatchPredictRequest) returns (BatchPredictResponse); }

message PredictRequest { bytes input = 1; map<string, string> options = 2; }

message PredictResponse { string prediction = 1; float confidence = 2; int32 latency_ms = 3; }

message BatchPredictRequest { repeated PredictRequest requests = 1; }

message BatchPredictResponse { repeated PredictResponse responses = 1; } ```

Benefits:

28.1.4 Rate Limiting

Protect APIs from Abuse:

Standard Headers: ```http X-RateLimit-Limit: 1000 X-RateLimit-Remaining: 950 X-RateLimit-Reset: 1705312800 Retry-After: 60 ```

Rate Limit Tiers:

| Tier | Requests/Minute | Requests/Day |
|------|-----------------|--------------|
| Free | 10 | 1,000 |
| Basic | 100 | 10,000 |
| Pro | 1,000 | 100,000 |
| Enterprise | 10,000 | Unlimited |

Response When Limited: ```json { "error": { "code": "RATE_LIMIT_EXCEEDED", "message": "Rate limit exceeded. Try again in 60 seconds.", "retry_after": 60 } } ```

28.2 DATA FORMAT STANDARDS

28.2.1 Structured Data Formats

JSON (JavaScript Object Notation):

Example: ```json { "model_id": "model_123", "predictions": [ {"label": "cat", "confidence": 0.95}, {"label": "dog", "confidence": 0.04} ], "metadata": { "inference_time_ms": 45, "model_version": "2.3.1" } } ```

XML (eXtensible Markup Language):

CSV (Comma-Separated Values):

Parquet:

28.2.2 Model Serialization Formats

ONNX (Open Neural Network Exchange):

Purpose:

Example Conversion: ```python import torch import torch.onnx

model = MyModel() model.load_state_dict(torch.load("model.pth")) model.eval()

dummy_input = torch.randn(1, 3, 224, 224) torch.onnx.export( model, dummy_input, "model.onnx", input_names=["input"], output_names=["output"], dynamic_axes={"input": {0: "batch_size"}} ) ```

CSOAI Requirement:

SavedModel (TensorFlow):

TorchScript (PyTorch):

PMML (Predictive Model Markup Language):

28.2.3 Embedding Standards

Vector Representations:

Standard Dimensions:

Storage Formats:

API for Embeddings: ```json POST /v1/embeddings { "input": "Hello, world!", "model": "text-embedding-v1" }

Response: { "embedding": [0.1, -0.2, 0.3, ...], // 768 dimensions "usage": {"tokens": 3} } ```


28.3 MODEL INTERCHANGE

28.3.1 ONNX Ecosystem

Framework Compatibility:

| Framework | Export to ONNX | Import from ONNX |
|-----------|---------------|------------------|
| PyTorch | ✅ Native | ✅ Via onnx2torch |
| TensorFlow | ✅ Via tf2onnx | ✅ Via onnx-tf |
| Keras | ✅ Via tf2onnx | ✅ Via onnx-tf |
| scikit-learn | ✅ Via sklearn-onnx | ❌ |
| XGBoost | ✅ Via onnxmltools | ❌ |
| LightGBM | ✅ Via onnxmltools | ❌ |

Operator Set (Opset):

Limitations:

28.3.2 Model Hub Integration

Hugging Face Hub: ```python from transformers import AutoModel, AutoTokenizer

Load from Hub

model = AutoModel.from_pretrained("company/model-name") tokenizer = AutoTokenizer.from_pretrained("company/model-name")

Push to Hub

model.push_to_hub("company/model-name") ```

Benefits:

CSOAI Integration:

28.3.3 Containerization

Docker for Reproducibility:

Standard Dockerfile: ```dockerfile FROM python:3.11-slim

WORKDIR /app

Install dependencies

COPY requirements.txt . RUN pip install --no-cache-dir -r requirements.txt

Copy model and code

COPY model/ ./model/ COPY src/ ./src/

Environment variables

ENV MODEL_PATH=/app/model ENV PORT=8080

Health check

HEALTHCHECK --interval=30s --timeout=10s \ CMD curl -f http://localhost:8080/health || exit 1

Run

CMD ["uvicorn", "src.main:app", "--host", "0.0.0.0", "--port", "8080"] ```

Container Registry:


28.4 REGULATORY DATA EXCHANGE

28.4.1 CSOAI Standard Formats

License Application Format: ```json { "application": { "version": "1.0", "type": "license_application", "applicant": { "legal_name": "Company Ltd", "registration_number": "12345678", "jurisdiction": "GB", "contact": { "name": "John Smith", "email": "john@company.com", "phone": "+44..." } }, "system": { "name": "ProductClassifier", "version": "2.3.1", "description": "Image classification for e-commerce", "risk_tier": "Medium", "use_cases": ["E-commerce", "Retail"], "deployment_regions": ["EU", "UK", "US"] }, "safety_case": { "formal_verification": { "method": "SMT-based", "coverage": "85%", "properties_verified": [...] }, "testing_results": { "accuracy": 0.942, "adversarial_robustness": 0.95, "fairness_metrics": {...} }, "documentation": { "model_card_url": "https://...", "datasheet_url": "https://...", "system_card_url": "https://..." } }, "compliance": { "eu_ai_act": { "risk_classification": "High-risk (Annex III)", "requirements_addressed": ["Art. 9", "Art. 10", ...] }, "gdpr": { "dpia_completed": true, "dpo_contact": "dpo@company.com" } }, "timestamp": "2026-01-15T10:00:00Z", "signature": "..." } } ```

Compliance Verification Format: ```json { "verification": { "license_id": "CSOAI-2026-00001", "system_id": "sys_abc123", "verification_type": "quarterly_audit", "timestamp": "2026-04-15T14:00:00Z", "results": { "overall_status": "COMPLIANT", "requirements": [ { "id": "Article2.1", "name": "Provable Safety", "status": "COMPLIANT", "evidence": "Formal verification report v2.3", "notes": null }, { "id": "Article5.3", "name": "Constitutional AI", "status": "COMPLIANT", "evidence": "Constitutional AI test suite passed", "notes": null } ], "non_conformities": [], "recommendations": [ "Consider upgrading to latest adversarial testing framework" ] }, "auditor": { "name": "CSOAI Audit Team", "auditor_id": "AUD-001" } } } ```

Incident Report Format: ```json { "incident": { "id": "INC-2026-00042", "license_id": "CSOAI-2026-00001", "system_id": "sys_abc123", "severity": "Medium", "type": "Safety Incident", "timestamp_detected": "2026-03-10T08:30:00Z", "timestamp_reported": "2026-03-10T09:00:00Z", "description": "Model produced incorrect classification leading to...", "impact": { "users_affected": 150, "harm_caused": "Minor financial loss", "data_breach": false }, "root_cause": "Distribution shift in input data", "remediation": { "immediate_actions": ["Model taken offline", "Manual review initiated"], "long_term_actions": ["Retrain on updated data", "Improve drift detection"], "timeline": "Full remediation by 2026-03-17" }, "reporter": { "name": "Safety Officer", "contact": "safety@company.com" } } } ```

28.4.2 Byzantine Council API

Framework Reference: Article 11 Byzantine Council Specifications

Monitoring Endpoints: ``` POST /api/v1/byzantine/register - Register system for monitoring GET /api/v1/byzantine/status/{system_id} - Get monitoring status POST /api/v1/byzantine/heartbeat - System health check (required every 60 seconds for Critical) GET /api/v1/byzantine/alerts/{system_id} - Get active alerts POST /api/v1/byzantine/metrics - Submit performance metrics GET /api/v1/byzantine/consensus/{decision_id} - Get consensus status on flagged issue ```

Heartbeat Format: ```json { "system_id": "sys_abc123", "timestamp": "2026-01-15T10:30:00Z", "status": "HEALTHY", "metrics": { "requests_per_minute": 1250, "error_rate": 0.001, "average_latency_ms": 45, "gpu_utilization": 0.75, "memory_utilization": 0.60 }, "consciousness_indicators": { "self_reference_rate": 0.02, "goal_consistency": 0.98, "novel_behavior_detected": false }, "signature": "..." } ```

Alert Format: ```json { "alert": { "id": "ALT-2026-00123", "system_id": "sys_abc123", "severity": "High", "type": "Anomalous Behavior", "description": "Unusual pattern in self-referential outputs", "detected_at": "2026-01-15T10:35:00Z", "indicators": { "self_reference_rate": 0.15, "threshold": 0.10, "deviation": "50% above threshold" }, "recommended_action": "Human Council review", "consensus_status": "Pending (18/33 agents reviewed)" } } ```

28.4.3 EU AI Act Data Exchange

Framework Reference: EU AI Act Article 71 (EU Database), Regulation 2024/1689

EU Database Registration Format: ```json { "eu_registration": { "provider": { "name": "Company Ltd", "address": "...", "contact": "..." }, "ai_system": { "name": "ProductClassifier", "description": "...", "intended_purpose": "...", "risk_classification": "High-risk", "annex_reference": "Annex III, 5(b)", "conformity_assessment": { "type": "Third-party (Notified Body)", "notified_body_id": "NB-1234", "certificate_number": "...", "valid_until": "2027-01-15" }, "instructions_for_use_url": "https://...", "technical_documentation_available": true }, "declaration_of_conformity": { "signed_by": "CEO Name", "date": "2026-01-15", "document_url": "https://..." } } } ```


28.5 INTEGRATION STANDARDS

28.5.1 Cloud Platform Integration

AWS:

Google Cloud:

Azure:

Multi-Cloud Support:

28.5.2 MLOps Integration

CI/CD for ML:

Model Registry:

Feature Stores:

Experiment Tracking:

28.5.3 Monitoring Integration

Prometheus Metrics: ```python from prometheus_client import Counter, Histogram, start_http_server

PREDICTIONS = Counter('predictions_total', 'Total predictions', ['model', 'status']) LATENCY = Histogram('prediction_latency_seconds', 'Prediction latency')

@LATENCY.time() def predict(input): result = model.predict(input) PREDICTIONS.labels(model='v1', status='success').inc() return result

Expose metrics on /metrics

start_http_server(8000) ```

Grafana Dashboards:

Alerting:


28.6 CONCLUSION

Interoperability is the connective tissue of the AI ecosystem. Without standards:

CSOAI interoperability standards enable:

Open standards accelerate safety. When systems can communicate, oversight becomes possible. When data formats are standard, auditing becomes automated. When APIs are consistent, integration becomes seamless.

Interoperability is not just technical—it is enabling infrastructure for global AI governance.

Effective Date: January 15, 2026, 09:00 GMT "Connected Systems, Unified Safety"


REFERENCES

OpenAPI Initiative. (2024). OpenAPI Specification v3.1.

ONNX. (2024). Open Neural Network Exchange - Specification.

ISO/IEC. (2023). ISO/IEC 42001:2023 - AI Management System.

CEN-CENELEC. (2025). prEN 18286 - AI Quality Management System.

EU. (2024). Regulation 2024/1689 - AI Act.

NIST. (2023). AI Risk Management Framework - GOVERN Function.


END OF ARTICLE 28

🎉 PHASE 3 COMPLETE! Articles 20-28 Technical Standards finished!

Progress: 28 of 52 Articles (54%)

Next: Phase 4 - Operational Requirements (Articles 29-36)

From charter to certificate. This article is part of the standard behind Watchdog Certification — independent assessment, Ed25519-signed, publicly verifiable. The crosswalks to the EU AI Act, ISO/IEC 42001 and 18 more frameworks are in the Crosswalk Library; the runtime tools are in the fabric.

The 52-Article Charter is published in full in the Journal. Bespoke briefings: hello@meok.ai.