Enterprise Trust Center

1. Trust Principles

At InvisibleAct, trust is foundational to enabling autonomous AI agents in mission-critical environments. Our platform is engineered around three non-negotiable principles:

1. Zero Customer Content Storage (Data Minimization): We do not inspect, ingest, or store prompt text, model completions, or customer application payloads.
2. Deterministic Pre-Execution Enforcement: Actions execute if and only if explicit policy authorization (ALLOW) is verified before tool execution.
3. Transparent Governance: We state clearly what is currently implemented, what is configurable, and what remains on our forward roadmap without exaggerated compliance claims.

---

2. Capabilities Breakdown

To provide complete clarity to security, privacy, and compliance teams, our capabilities are classified across four clear lifecycle tiers:

2.1 Privacy Architecture & Data Minimization

• IMPLEMENTED Metadata-Only Gatekeeping: Evaluates actions based strictly on structured metadata (tenant_id, agent, vendor, action, amount, tool, data_access).
• IMPLEMENTED Zero Prompt / Content Ingestion: Customer chat history, document attachments, and LLM responses are never sent to or stored by InvisibleAct.
• IMPLEMENTED Production Log Sanitization: Standard output container logs strip sensitive tenant identifiers, session tokens, action parameters, and trajectory history.
• NOT YET IMPLEMENTED Automated Unstructured PII Scrubbing: Because InvisibleAct does not ingest unstructured text or customer documents, prompt-level PII scrubbing pipelines are neither implemented nor required.

2.2 Pre-Execution Authorization & Policy Enforcement

• IMPLEMENTED Three-State Decision Engine: Pre-execution enforcement returns ALLOW, REVIEW, or BLOCK. Zero tools execute on REVIEW or BLOCK.
• IMPLEMENTED Registered Agent Verification: Evaluates agents against an isolated tenant agent directory. Unregistered or disabled agents are blocked.
• IMPLEMENTED Vendor Trust Verification: Restricts agent operations to tenant-approved model vendors and providers.
• IMPLEMENTED Exploit Trajectory Detection: Detects conflicting multi-step action sequences (such as invoice deletion followed by funds transfer) and suspends execution.
• IMPLEMENTED Cumulative Spending Ceilings: Automatically tracks multi-action cumulative spend across active sessions and blocks actions exceeding defined ceilings.
• IMPLEMENTED Single-Use Human Approvals: High-risk actions suspended for human review can be approved by authorized operators; once consumed, approvals cannot be replayed.
• IMPLEMENTED Model Context Protocol (MCP) Adapter: Thin, secure proxy intercepting MCP tools/call requests with zero-bypass pre-execution authorization.

2.3 Tenant Isolation & Access Governance

• IMPLEMENTED Row-Level Database Partitioning: All tenant data is strictly partitioned by tenant identifier at the database query layer.
• IMPLEMENTED Cryptographic Tenant Binding: API keys and dashboard sessions are bound to a single tenant. Cross-tenant access is rejected with HTTP 403 Forbidden.
• IMPLEMENTED Role-Based Access Control (RBAC): Five discrete roles (machine_agent, tenant_admin, approval_reviewer, auditor, master) enforce strict least privilege.
• IMPLEMENTED One-Way API Key Hashing: Plaintext keys are shown only once upon creation and stored exclusively as one-way SHA-256 hashes.
• IMPLEMENTED Session Lifecycle Governance: 12-hour absolute TTL, 30-minute idle timeout, and instantaneous server-side logout invalidation.
• NOT YET IMPLEMENTED Physical Database Isolation: A single shared database schema is utilized with logical row-level tenant partitioning rather than dedicated physical databases per tenant.

2.4 Auditability, Retention & Deletion

• IMPLEMENTED Verifiable Audit Trails: Every authorization evaluation is logged in an immutable audit ledger with request UUID, millisecond timestamp, risk score, and actor attribution.
• IMPLEMENTED Reviewer Attribution: Human approval events record the authenticated reviewer's identity rather than a static placeholder.
• CONFIGURABLE Audit Log Retention Window: Configurable retention periods enable organizations to establish customized data lifecycle schedules.
• IMPLEMENTED Age-Based Retention Cleanup: Automated retention routines remove expired audit records while preserving recent active history.
• IMPLEMENTED Administrative Tenant Deletion: An authenticated administrative tenant deletion process purges all audit logs, session states, policies, vendors, agents, API keys, and sessions for a specified tenant without cross-tenant impact.
• NOT YET IMPLEMENTED Public Self-Service GDPR Deletion Portal: Tenant data deletion is currently an authenticated administrative function executed via Master Administrator credentials.
• NOT YET IMPLEMENTED Built-in Automated Background Cron: Periodic audit retention purging requires an external scheduler (such as a Kubernetes CronJob) to invoke the cleanup API.

2.5 Infrastructure & Communications Security

• IMPLEMENTED Transport Layer Security (TLS): All public endpoints operate over TLS/HTTPS.
• IMPLEMENTED Database TLS Enforcement: Managed PostgreSQL connections enforce TLS (sslmode=require).
• IMPLEMENTED Fail-Closed Database Connection: If cloud PostgreSQL is unreachable, the gateway fails closed and strictly refuses unencrypted fallback in production.
• IMPLEMENTED HTTP Security Headers: Enforces nosniff, DENY frame options, XSS blocking, and referrer policy while stripping server software headers.
• IMPLEMENTED Production Error Sanitization: Prevents internal stack trace or file path disclosures.
• CONFIGURABLE Rate Limiting: Sliding-window rate limiting on public endpoints (/leads, /auth/session) to defend against brute force.

---

3. Compliance & Certification Roadmap

InvisibleAct is committed to meeting the highest international security and privacy benchmarks. To ensure complete integrity, our current certification status and roadmap are presented below:

---

4. Contact & Security Reporting

For enterprise security reviews, customer vendor assessments, or privacy requests:

• Security & Privacy Team: [email protected]
• Vulnerability Disclosure: [email protected]
• Sales & Pilot Inquiries: [email protected]