Client Data Processing Policy & Sovereign AI Framework

Rochon is architected from the foundation up for zero-egress, perimeter-contained enterprise intelligence. We reject the commodity paradigm of routing sensitive corporate prompts and telemetry to third-party public cloud APIs.

All models, weights, vector indexes, and reasoning pipelines engineered by Rochon are deployed directly into infrastructure owned and controlled exclusively by the Client: on-premise physical GPU racks (RTX/L40S), client-managed private cloud VPCs (AWS, Azure, GCP), or local embedded edge silicon (NVIDIA Jetson, NXP, Silicon Labs). External network calls are eliminated or locked to verified, client-audited network boundaries.

Under no circumstances does Rochon use Client-provided datasets, operational logs, schema definitions, internal documentation, or runtime inputs to train, pre-train, fine-tune, or evaluate public models or proprietary lab models.

Client data is permanently isolated within the Client's tenant or air-gapped sandbox. Weights generated through fine-tuning (LoRA, DoRA, Full Parameter, DPO, RLHF) are cryptographically stored inside Client boundaries and remain the unencumbered, sovereign property of the Client.

In the execution of AI engineering and deployment services, Rochon operates under well-defined international data protection classifications:

• Client as Data Controller / Data Fiduciary: The Client maintains sole authority over the purposes, legal grounds, and means of processing personal data, retaining full status as Data Controller under EU/UK GDPR, UAE PDPL, and KSA PDPL, and Data Fiduciary under the Indian DPDPA 2023. • Rochon as Data Processor: Rochon acts strictly as a Data Processor / Service Provider under the Client's documented written instructions and binding Data Processing Addenda (DPAs). • Sub-Processors: Rochon does not engage third-party sub-processors to access, store, or process Client data without explicit, prior written authorization from the Client.

Rochon's architecture guarantees compliance with the world's most demanding regional data residency regulations:

• India (DPDPA 2023 & MeitY): Client data processed in India can remain strictly within domestic sovereign boundaries (e.g., local on-premise servers or Indian hyperscaler zones in Mumbai/Hyderabad), fulfilling all Central Government cross-border transfer requirements. • Kingdom of Saudi Arabia (KSA PDPL & SDAIA): Complete support for National Data Governance Standards, ensuring enterprise and governmental records do not egress the Kingdom without requisite approvals and statutory transfer mechanisms. • United Arab Emirates (UAE PDPL & DIFC/ADGM): Turnkey deployments into sovereign UAE cloud availability zones or on-premise data centers, satisfying federal and free-zone cross-border data protection mandates. • European Union (GDPR Chapter V & Schrems II): Elimination of foreign cloud surveillance exposure by deploying air-gapped, on-premise inference engines, providing complete legal certainty under CJEU Schrems II rulings and EU Standard Contractual Clauses. • United States: Zero-egress architectures designed to satisfy HIPAA Security Rule specifications for healthcare, GLBA / FINRA criteria for financial institutions, and NIST SP 800-171 controls for the defense supply chain.

When deploying AI to industrial machinery, autonomous robotics, medical devices, or remote microgrids, Rochon constructs fully autonomous, air-gapped runtime binaries:

• Self-Contained Inference: Run-time engines (built on TensorRT, ONNX Runtime, NXP eIQ, vLLM, or llama.cpp) require zero active internet connectivity, license validation handshakes, or cloud heartbeat pings to execute inference. • Physical Attack Surface Mitigation: Edge memory layouts are hardened with memory encryption, secure boot validation, and tamper-resistant model quantization, preventing unauthorized extraction of weights.

Rochon implements defense-in-depth security measures to protect client environments during all engineering stages:

• Cryptographic Standards: High-grade AES-256 encryption at rest for weights, cache tiers, and vector embeddings; TLS 1.3 encryption with strict cipher suites for all intra-cluster communications. • Zero-Trust Access Control: Role-Based Access Control (RBAC), multi-factor authentication, and integration with Client identity providers (SAML 2.0, OpenID Connect, Okta, Azure AD). • Immutable Audit Logging: Real-time cryptographic logging of all model routing decisions, token accounting, and administrative access, streaming directly to Client-managed SIEM log repositories.

Upon completion, expiration, or termination of an engineering engagement, Rochon executes a formal data sanitization workflow:

• Client Election: At Client's written election, all client datasets, synthetic samples, intermediate checkpoints, and validation logs are either securely transferred to Client or permanently eradicated. • Sanitization Standard: Deletion is performed following NIST Special Publication 800-88 Revision 1 (Guidelines for Media Sanitization) standards. • Certificate of Destruction: Rochon delivers a signed, formal Certificate of Destruction within fourteen (14) calendar days of verified erasure.

We maintain proactive transparency and collaborative verification:

• Audit Rights: Clients or their independent third-party auditors are entitled upon reasonable prior written notice to inspect deployment architectures, review cryptographic configurations, and verify compliance with this Policy. • Incident Escalation: In the event of a verified cybersecurity vulnerability or breach affecting client systems under our active management, Rochon will notify Client without undue delay and within statutory windows (including CERT-In 6-hour mandates, GDPR 72-hour notifications, and DPDPA requirements). • Regulatory Cooperation: Rochon commits to providing technical documentation and architectural affidavits to support Client regulatory filings and compliance certifications before national authorities.