Skip to content
Agent Trust Cloud

Trust center

Security infrastructure has to hold itself to the standard it sells

Agent Trust Cloud sits in a privileged position, so its own privacy posture, tenancy model, cryptography and development lifecycle are part of the product — and part of your buying decision.

Check the evidence-backed security delivery status: 8 controls live and 9 planned.

PLATFORM SECURITY

Privacy, Retention & Data Residency

Agent telemetry may contain prompts, customer records, proprietary code and regulated information. Privacy controls must be configurable from day one, including the ability to operate with metadata-only telemetry for sensitive customers.

What we build

  • Per-tenant and per-data-class retention periods.
  • Regional data residency options for control-plane data where commercially justified.
  • Prompt/content capture modes: full, redacted, hashed, metadata-only.
  • Deletion/export workflows for supported privacy requests and customer offboarding.
  • Sensitive-field classification and minimization.

How it is implemented

  • Keep operational metadata separate from content payloads.
  • Default to not storing secrets; aggressively redact credentials.
  • Document subprocessors and cross-region flows.
  • Use customer-managed keys as an enterprise option where needed.

PRIVACY BY DESIGN

Collect the least content needed to secure and explain the agent workflow.

PLATFORM SECURITY

Multi-Tenant Isolation

As a security vendor, a cross-tenant data leak would be existential. Tenant isolation must be designed across identity, database access, object storage, caches, queues, analytics and support tooling—not assumed from application-level filters alone.

What we build

  • Tenant-scoped authorization on every data-access path.
  • Strong database row/partition controls and service-layer checks.
  • Separate object-storage namespaces and encryption context.
  • Tenant-aware queues, caches and search indexes.
  • Privileged support access with approval and auditing.

How it is implemented

  • Automated negative tests for cross-tenant access.
  • Never trust tenant IDs supplied by unverified client inputs.
  • Use tenant context propagated from authenticated identity.
  • Design enterprise dedicated/regional options without forking the product.

NON-NEGOTIABLE

Security infrastructure cannot ask customers to accept weak multi-tenancy.

PLATFORM SECURITY

Cryptography & Key Management

Cryptographic design should support secure service identity, signed policy bundles, protected secrets and evidence integrity. Avoid custom cryptography; use mature cloud/HSM/KMS primitives and documented rotation procedures.

What we build

  • TLS everywhere; mTLS for sensitive service/gateway paths.
  • Envelope encryption for sensitive data stores.
  • KMS/HSM-backed signing keys for tokens, policy bundles and evidence manifests.
  • Per-tenant key options for enterprise requirements.
  • Key rotation, revocation and disaster-recovery procedures.

How it is implemented

  • Separate encryption keys from application data and least-privilege KMS access.
  • Track key version in signed/ encrypted artifacts for historical verification.
  • Automate rotation and alert on overdue keys/certificates.
  • Never hard-code secrets or private keys in code/build artifacts.

RULE

Use cryptography to prove identity and integrity; do not invent cryptographic algorithms.

PLATFORM SECURITY

Secure SDLC & Product Security

Customers will evaluate Agent Trust Cloud as critical security infrastructure. Product security must therefore be a company capability, not a checklist added before enterprise sales. Build controls into development, release and operations from the first production version.

What we build

  • Threat modeling for gateway, policy, identity, connectors, admin and evidence paths.
  • SAST, dependency/SBOM, secrets scanning and infrastructure-as-code scanning in CI.
  • DAST/API security testing and regular penetration tests.
  • Protected branches, reviewed production changes and signed builds where feasible.
  • Vulnerability intake, severity SLAs and coordinated disclosure policy.

How it is implemented

  • Maintain separate dev/test/prod accounts and secrets.
  • Create security regression tests for tenant isolation and policy bypass.
  • Log privileged production access and use just-in-time elevation.
  • Prepare SOC 2 readiness practices as enterprise traction develops.

TRUST SALES

The product’s own security program is part of the customer buying decision.

Customer review facts

Data handling and supplier transparency

Subprocessors and AI providers

  • Railway hosts the application; Neon hosts PostgreSQL; Microsoft 365 delivers operational email; Stripe processes payment when enabled.
  • Configured AI inference is routed through the governed provider layer. Provider, model and usage are recorded; customer content is not used for model training by Agent Trust Cloud.
  • A customer review is required before enabling any connector or provider that will receive its data.

Location, retention and requests

  • Deployment and database regions are confirmed in the order form before production onboarding; cross-border transfers require an approved mechanism.
  • Operational and evidence retention is set by the customer policy and contractual schedule; deletion and legal-hold exceptions are recorded.
  • Request the DPA, security material or subprocessor details at [email protected]. Data-protection questions use the same address.

The DPA is currently a draft for legal review and is not published on this site.

Where the product uses AI, what that feature sees and how to turn it off: how we use AI. Agent Trust Cloud does not currently process PHI under a BAA; see HIPAA and AI agents.