Home/Platform/Security & Compliance

The page compliance sends to risk.

A capability of frIdA.

Data residency, encryption posture, access controls, audit surface, subprocessor disclosure, and regulatory alignment. Everything a compliance committee needs to evaluate Saptiva AI against its existing control framework - stated plainly, including what we do not yet have.

Document versionv2.3
Last reviewed2026-Q3
Questionssecurity@saptiva.com
Security posture

Security is a property of the deployment, not a vendor claim.

The security posture a customer experiences depends on how their deployment is configured against their policy - not on a fixed property of the platform. A customer running in air-gapped mode has a materially different security envelope than a customer running in public cloud. Both are valid. The platform is the same in each case; the controls that bind it are different.

Everything on this page describes capability. Which capabilities are engaged in a specific deployment is determined by the customer's configuration, applied by frIdA, and reflected in the audit record. Compliance committees evaluating Saptiva AI should read this document alongside the specific deployment configuration proposed for their environment - not as a substitute for it.

Data residency

Where data lives, runs, and returns.

Residency is enforced by architecture, not by trust. Each data class is assigned the environments it may run in. frIdA routes accordingly and records the route. The customer defines the classes.

Mexico
In-country compute

Residency satisfied by architecture. In-country regions and on-premises, inside Mexico.

Central America
Regional

In-market deployment at MultiMoney. On-premises and private cloud.

Air-gapped
Zero external connectivity

Air-gapped deployments for critical infrastructure and the most sensitive financial workloads.

For deployments where residency is a defining requirement, we publish the physical region, the legal jurisdiction, and the contractual residency commitment in the customer agreement. The claim on this page is not the commitment. The commitment is in the MSA.

Encryption & keys

Cryptographic posture.

ControlPosture
Data in transitTLS 1.3 for all external traffic. Mutual TLS for service-to-service inside the deployment. No unencrypted network paths in production.
Data at restAES-256 for all stored data. Per-tenant key scoping. Storage-layer encryption is additive to application-layer encryption for sensitive fields.
Key managementCustomer-held keys supported in in-country, private, and on-premise deployments. HSM integration available. In public cloud mode, keys managed via the cloud provider's KMS with customer-controlled policies.
Key rotationAutomated rotation on a configurable schedule. Default is 90 days for data-at-rest keys. Audit record captures every rotation event.
Model weightsModel weights are treated as customer-scoped data where applicable. Fine-tuned customer models never leave the customer's deployment environment.
Model trainingCustomer Data is never used to train, fine-tune or evaluate any model, ours or a model provider's. Not as an aggregate, an embedding, or a derived dataset that outlives the deployment. Where a customer fine-tunes on their own data, the resulting weights are the customer's and stay in their environment.
Access controls

Who can do what, to which data.

Identity

SSO and directory integration

SAML 2.0 and OIDC. Integrates with the customer's directory (Active Directory, Okta, Azure AD). Saptiva AI does not maintain a separate identity provider for production users of customer deployments.

RBAC

Role-based access controls

Roles defined in the customer's directory and configuration. frIdA enforces role → capability mapping on every dispatch. Access decisions appear in the audit record with the role and justification captured.

Least privilege

Capability-scoped sessions

Application access is scoped to the specific capability needed for the task. A credit officer's copilot session does not see KYC data unless explicitly permitted.

MFA

Multi-factor enforced

MFA required for administrative access and for production workloads touching regulated data. Enforcement tied to the customer's directory policy.

Privileged access

Saptiva AI operator access

Saptiva AI personnel do not have standing access to customer production data. Operational intervention requires customer-authorized, time-bounded access tied to a specific incident or work order, with full session audit.

Session governance

Session timeouts and re-auth

Configurable by the customer. Defaults follow banking-grade expectations. Re-authentication required for privileged operations regardless of session state.

Audit surface

The record is the deliverable.

Every frIdA dispatch produces a signed, immutable audit record. The record captures what model ran, on what data, under which configuration version, authorized by which identity, and with what outcome. It is the object a regulator, internal audit, or post-incident investigation examines - not a log file, but a structured record.

Regulatory alignment

The frameworks we deploy under.

We name an authority here only where a workload is running under it today. Everything else is architecture, and architecture is on the rest of this page.

AuthorityDeployment posture
CNBV (Mexico · Banking)Production deployments active. Residency, audit and access controls satisfy the expectations reviewed by customer compliance committees.
CNSF (Mexico · Insurance)Production deployments active. Document processing and customer operations under the customer's compliance framework.
Everywhere elseThe customer configures the routes. frIdA applies them and records every run. We do not maintain a list of authorities, because the list was never ours. If your supervisor is not named above, that is not a gap. It is the design.
Certifications

What we have. What we don't yet.

Honest posture

Saptiva AI does not yet hold a completed SOC 2 Type II attestation. We are in the formal readiness and assessment phase with a qualified assessor. We expect completion in the current fiscal year and will publish the attestation date when it is signed.

ISO 27001 is in scope for the following fiscal year. We intend to pursue it because our enterprise customers ask for it - not because it materially changes our production security posture, which is already governed by the controls on this page.

Policy authoring inside the platform and policy-based routing - a rule written once and applied on every route - ship next. Until they ship, routing is configured by your operators and every route is recorded. We say so here so it does not surface at procurement.

If your compliance framework requires a completed SOC 2 or ISO 27001 attestation before a vendor can be onboarded, we can share our current readiness artifacts under NDA and connect your compliance team with our security lead. We prefer to say so up front rather than have the conversation surface at procurement.

We hold the following operational certifications and alignments:

Subprocessors

Who touches the data, beyond us.

The subprocessor list depends on the deployment mode. Air-gapped deployments have no subprocessors. Public cloud deployments inherit the cloud provider as a subprocessor. In-country and on-premise deployments may or may not involve additional subprocessors depending on the customer's configuration.

For each customer engagement, we publish the active subprocessor list in the DPA and notify the customer in advance of any change. The canonical subprocessor list is available under NDA to prospective customers; active customers receive it as part of their DPA.

CategoryTypical subprocessors
Compute infrastructurePublic cloud deployments: AWS, Microsoft Azure, or Google Cloud, depending on customer selection. In-country deployments: the in-country operator of the region. On-premise: none.
Hardware distributionHPE, Dell, and NVIDIA through distribution partnerships for on-premise and hybrid deployments. These parties distribute hardware; they do not process customer data.
Operational supportForward Deployed Engineers and operational staff employed by Saptiva AI. Third-party operational subprocessors are not routine; any use is disclosed to the customer in advance.
Model providersDepends on deployment configuration. Customers may run open-weight models with no external model subprocessor. Where commercial models are used, the provider appears as a subprocessor for that specific workload.
Incident response

When something goes wrong.

Our incident response posture is built around timely customer notification, joint technical response, and post-incident transparency. The specifics are governed by the MSA and DPA for each customer.

Reporting a concern

Direct line to security.

For security concerns, vulnerability reports, compliance questions from prospective customers' risk teams, or responsible disclosure - use the address below. It is monitored by our security lead and routed appropriately.

We will not pursue or support a claim against a researcher who acts in good faith. The conditions that define good faith are set out once, in section 011 of the Terms of Use, so that there is a single version of them.

Security contact

For compliance questionnaires and DPA review from prospective customers, the same address applies.

Contact

Your compliance committee will have questions. So will your CISO.

A Forward Deployed Engineer and our security lead can jointly walk your compliance and security teams through the deployment proposed for your environment - tied to your specific regulatory framework and your existing control posture. Reply within 48 hours.