🧠 AI Data Governance & Edge Processing Policy
Last updated: November 14, 2026This Artificial Intelligence Data Governance Policy sets forth the technical, contractual, and ethical parameters governing the processing of enterprise data across solutions architected by SoftMena, C.A. (hereinafter, "SoftMena"), including our perimeter computer vision suite (EyeSentinelAI), multi-agent cognitive pipelines, and asynchronous B2B integrations.
1. Technical Roles: Data Controller vs. Data Processor
- The Corporate Client as Data Controller: The client retains exclusive ownership and control over its on-premise infrastructure, closed-circuit television (CCTV) cameras, video management systems (VMS), proprietary databases, and enterprise ERP records. The client is solely responsible for establishing the lawful basis for data collection (e.g., employment contract, legitimate security interest, or regulatory compliance) and for deploying visible physical/digital surveillance notices to employees, visitors, and customers.
- SoftMena, C.A. as Data Processor (Technical Orchestrator): SoftMena operates strictly as a technology orchestrator and technical processor executing instructions defined in the Master Services Agreement (MSA) and Statement of Work (SOW). SoftMena does not determine the purposes of client data processing, does not monetize client datasets, and does not claim ownership over client video streams or business data.
2. AI Training Principles: Zero Public / Cross-Tenant Training vs. Dedicated Fine-Tuning
SoftMena establishes a transparent architectural separation between public foundational models and dedicated client intelligence:
- Zero Public / Cross-Tenant Training: Neither SoftMena nor any underlying pipeline utilizes client surveillance footage, RTSP streams, proprietary datasets, codebases, documents, or operational telemetry to train, retrain, or align public or shared multi-tenant foundation models (including OpenAI, Anthropic, Google DeepMind, or open-source community repositories). Client operational data is never utilized to benefit other clients or third-party systems.
- Dedicated Fine-Tuning and Isolated Vector Indexing: When explicitly authorized and contracted by the client (such as bespoke asset recognition, custom object detection weights, or enterprise Retrieval-Augmented Generation / RAG workflows), SoftMena performs dedicated fine-tuning and isolated vector embeddings indexing (e.g., Qdrant vector databases, FaceNet feature extractors, or local embedding engines). All resulting weights, vector indices, and specialized adapters are the exclusive property of the client, remain confined within the client's private dedicated runtime environment, and are completely isolated from external tenants.
3. Hybrid Edge / VPC Topology, Real-Time In-Memory Processing, and Rolling Retention
Our flagship vision platform, EyeSentinelAI, and operational neural pipelines are architected with flexible high-security topologies:
- Edge and Dedicated Private Cloud (VPC) Deployment: EyeSentinelAI operates flexibly across local edge nodes (industrial neural appliances, NVIDIA Jetson modules, or on-premise edge servers isolated in CCTV VLANs) or dedicated private cloud servers / Virtual Private Clouds (such as dedicated infrastructure in Hetzner or private cloud tenancies) interconnected via end-to-end encrypted tunnels (IPsec / WireGuard VPN).
- Real-Time Transient In-Memory Processing: Video frames and streaming telemetry are processed in volatile RAM buffers in real time with high concurrency and low latency. Transient frames are wiped from volatile memory immediately after neural feature extraction.
- Minimalist Encrypted Telemetry: Systems generate and transmit only lightweight, structured JSON metadata (perimeter intrusion alerts, bounding boxes, classification, and event timestamps) over modern encrypted protocols in transit and at rest.
- Rolling Buffer Storage & Secure Sanitization: Incident video clips and evidence snapshots are retained only when contractually required, operating under a cyclic rolling buffer (FIFO overwrite) aligned with the client's agreed Service Level Agreement (SLA). Upon contract termination, all telemetry, weights, and configurations on managed hardware are securely purged and sanitized.
4. Corporate AI Autonomous Agents and RAG Governance
SoftMena designs and orchestrates corporate multi-agent workflows and cognitive automation under strict enterprise isolation:
- Private Instances and Zero Data Retention APIs: Corporate AI agents operate exclusively on private self-hosted models or enterprise-tier API instances governed by strict Zero Data Retention (ZDR) and no-training commercial agreements.
- Boundary Protection: Prompts, internal knowledge bases, enterprise ERP queries, and generated agent responses are never logged for third-party model training and remain strictly within the corporate security boundary.
5. Mandatory Human Oversight (Human-in-the-Loop) and Prohibited Uses
- Human-in-the-Loop Governance: Automated neural alerts and computer vision outputs serve exclusively as operational decision-support systems. No critical action affecting employment status, physical freedom, legal rights, or disciplinary measures may be executed solely on automated algorithmic inference without qualified human verification.
- Strictly Prohibited Uses: SoftMena strictly prohibits the use of its technology for unlawful biometric mass surveillance, clandestine facial harvesting without explicit legal grounds, discriminatory profiling based on protected characteristics, or corporate espionage.
6. Governance Contacts and Cross-Policy Framework
For security compliance, auditing requests, or AI governance clarifications:
- AI Governance & Security: admin@softmena.com
- Direct Corporate Line: +58 412-8893110
- Registered Office: Piso 1, C.C. San Miguel Arcángel, Local 11, Calle Negro Primero, Palo Negro 2117, Aragua, Venezuela.
Related legal documentation: