Trust Center

Intent Attestation & Governance Guarantees

• Authorized Intent Record (AIR): Lens generates a deterministic record of the user’s stated intent — composed of an abstract Motive, explicit Scope terms extracted verbatim, and a structural Boundary token. AIR is a pre-execution evidence artifact used by SOC2, HIPAA, and GRC teams to compare user-stated intent with downstream system behavior.
• Least-Privilege Scope Extraction: Lens captures only the explicit terms and entities stated in the input. No inference, summarization, or enrichment is performed; Scope reflects exactly what the user provided.
• Pre-Execution Boundary Token: Each AIR includes a static, structural Boundary token that customers may use as a contract point for downstream policy engines. Any deviation from Scope can be flagged, blocked, or audited by customer-side logic.
• Audit & Forensics Baseline: AIR serves as a stable comparison point between what the user stated and what the agent produced. This supports SOC2 activity logging, HIPAA minimum-necessary validation, incident analysis, and multi-agent oversight.
• Rogue-Model Detection: Any downstream action outside the Scope extracted by Lens represents a deviation from the attested intent boundary. Customers may use this as a signal for alerts, quarantines, or automated controls.
• Deterministic & Reproducible Output: Lens is designed for invariance. Given the same input, it produces the same AIR every time, ensuring auditability and consistent governance behavior.
• Attestation-Only Architecture: Lens performs no policy enforcement, access control, or decision-making. Lens never approves, denies, or executes actions. Customers retain full authority over their agents, policies, and workflows.

Lens Guarantees

• Zero Content Retention: Null Lens never stores, logs, caches, or persists user inputs or outputs. No prompt or response content enters any database, log file, monitoring tool, or long-term memory location.
• Stateless by Design: Each request is processed in isolation with no cross-request linkage. No data is carried, shared, or reused across calls.
• Application-Layer Only: Null performs no inference, no enrichment, no search, no retrieval, and no downstream execution. We only generate a deterministic M·S·B intent record.
• Deterministic Output: Given identical input, Null Lens produces structurally and semantically identical output. No randomness or drift is introduced at the application layer.
• Vendor-Bound Persistence: All physical hosting, compute, storage, logging, and network controls are the responsibility of our audited vendors (RunPod, Vercel, Supabase, Clerk). Null introduces no additional persistence surface.
• Private GPU Isolation (Enterprise): Enterprise tenants receive a dedicated GPU instance and non-discoverable private endpoint. No shared runtime or memory space is used.
• Outbound IP Minimization: Null returns only the Motive · Scope · Boundary schema. No internal prompts, system instructions, or model reasoning paths are ever emitted.

Responsibility Matrix

This matrix defines the exact boundary of responsibility across Null Lens, our audited infrastructure vendors, and customer systems. It reflects our stateless, application-layer-only architecture.

Infrastructure

• Core inference is executed on RunPod-managed GPU infrastructure. RunPod is an independent provider with its own security processes and third-party attestations; RunPod is responsible for the underlying orchestration and physical hosting. We maintain application-level monitoring and recovery automation on top of this base.
• In the default (shared) offering, requests are isolated at the process level and inference is stateless — no cross-request persistence is performed by our inference service.
• The front end and management dashboard are hosted on Vercel. Vercel handles static assets, CDN, and edge routing — it does not perform model inference for Null Lens.
• Organization records, billing metadata, and API key tracking are stored in Supabase. We store only account-level metadata there (no prompt or response content).
• All network traffic between client and service is encrypted in transit (HTTPS/TLS). Data at rest in vendor storage (where applicable) is protected by standard encryption (vendor-managed AES-256).

Security & Compliance

• Authentication and identity are handled by Clerk (third-party provider). Clerk manages auth, sessions, and user lifecycle; owners can rotate and revoke API keys at any time.
• We intentionally follow a least privilege approach at the application layer and require organization-scoped API keys for programmatic access.
• Null Lens itself has not undergone its own SOC 2 audit. However, all core vendors in our stack maintain independent compliance attestations: Clerk, Supabase, and Vercel are SOC 2 Type II certified, while RunPod holds SOC 2 Type I (Type II in progress). Null Lens builds upon the audited security controls of these providers and implements additional application-level controls to maintain data integrity and isolation.
• Lens is designed for deterministic governance, not data aggregation — every request is treated as an isolated transaction.
• We perform automated vulnerability scanning and periodic manual code review before production releases. Critical fixes are handled with priority patches and incident processes.
• Risk transparency: we rely on third-party infrastructure providers for availability, physical security, and certain controls. If a vendor has an outage or incident, recovery and post-incident actions will follow that vendor’s published processes. We coordinate and communicate incidents to customers where appropriate.

Data Governance

• Null Lens does not store or train on user input or output content in production. Inference occurs in-memory and remains ephemeral by design.
• Minimal operational metadata is recorded only for billing and reliability monitoring (organization ID, request latency, status codes). Prompt text and model responses are never logged or persisted.
• Customers retain full ownership of their data. Null Lens does not claim, reuse, or analyze inference content in any form.
• Audit & Retention: Null Lens does not provide long-term data retention or content logging. Customers who require audit trails or archival storage should implement these controls within their own infrastructure or middleware.
• PII & Regulated Data: Customers should avoid transmitting sensitive personal, health (PHI), or financial data unless their legal team has confirmed that use case and formally requested the necessary compliance addenda. Data Processing Addendums (DPAs) or Business Associate Agreements (BAAs) are executed only upon customer request and when legally required.

Reliability & SLA

• Null Lens operates on RunPod (compute) and Vercel (web edge) infrastructure. Each vendor maintains its own service-level targets and compliance controls, including multi-region redundancy and failover.
• Null monitors the application layer for uptime, latency, and error rates, and provides best-effort operational support built on top of these vendor platforms. Infrastructure-level availability remains subject to the vendors’ respective SLAs.
• For shared deployments, uptime expectations align with RunPod’s published infrastructure targets (typically 99.9% or higher). For private enterprise instances, custom operational commitments can be defined by contract.
• Typical observed request latency is approximately 1–5 seconds, varying with model size, queue depth, and regional load.
• In the event of a vendor outage, Null coordinates customer communication and incident reporting, following the vendor’s restoration and failover process. Post-incident summaries are available upon request.

Deployment Models

• Shared Instance — default multi-tenant offering. Process-level isolation and ephemeral inference without persistent content storage.
• Private GPU Instance (Enterprise) — we can provision an isolated GPU pod and private API endpoint for enterprise customers under contract. This reduces shared surface area and offers stronger operational separation.
• Self-Hosted / Air-Gapped — not offered by default. To protect proprietary model logic and licensing boundaries, Null Lens does not distribute its inference engine or model weights outside controlled infrastructure. For regulated environments requiring stronger isolation, we provide dedicated private GPU instances within audited RunPod facilities under NDA and enterprise agreement.

Legal & Contact

• Compliance documents, including Data Processing Addendum (DPA) or Business Associate Agreement (BAA), can be provided to enterprise customers upon request and subject to NDA. These are handled case-by-case based on the customer’s regulatory requirements.
• For vulnerability disclosure, compliance requests, or enterprise inquiries, contact us at support@null-core.ai. For urgent security issues you may request an NDA before sharing sensitive material.
• We are committed to transparency: vendor compliance documentation (Clerk, Supabase, Vercel, RunPod) can be reviewed during our enterprise diligence process or by request.
• © 2026 Null Technologies Pte. Ltd.