Subprocessors and Third-Party Data Sharing
Short answer
ProvenanceOne is built on cloud infrastructure and uses a small set of confirmed third-party services. When tenant workspaces configure AI model integrations, workflow and agent data is sent to the AI providers those tenants select (Anthropic, OpenAI, Google, Azure OpenAI). Stripe processes billing. A complete, audited subprocessor register with GDPR Article 28 coverage has not been confirmed. This page states what is confirmed and what must be verified before procurement.
What customers need to know
All data processed by ProvenanceOne travels through cloud infrastructure (primary subprocessor). When an agent step in a workflow invokes an AI model, the relevant context — system prompt, workflow context, tool descriptions, tool call results, agent memory — is sent to the AI provider configured for that agent. ProvenanceOne does not control the data handling or retention practices of those AI providers. Tenants are responsible for reviewing each provider's enterprise data handling documentation before configuring that provider in their workspace.
Warning: If your data classification or regulatory requirements restrict which third parties can process your data, audit the AI providers your tenants configure before allowing production workloads to run.
Section 1: Confirmed infrastructure subprocessors
Cloud infrastructure
ProvenanceOne runs on enterprise-grade cloud infrastructure. The platform uses the following categories of managed services:
| Service category | Purpose |
|---|---|
| Serverless compute | Workflow step execution and background processing |
| Managed database | Primary data store for workflow state, audit events, and workspace data |
| Object storage | Datastores, workflow artefacts, and large message payloads |
| Workflow orchestration | Durable execution of multi-step workflow runs |
| API management | API endpoint management and TLS termination |
| Identity and authentication | User authentication and workspace identity management |
| Secrets management | Credential and secret storage for connections and workspace secrets |
| Transactional email | Approval notifications, audit digests, and member invitations |
| Encryption key management | Key management for secrets, audit event MACs, and approval tokens |
| Container hosting | MCP servers in hosted execution mode |
| Event routing | Internal event routing and scheduled triggers |
| Message queuing | Bus subscriptions and background processing |
| CDN and edge delivery | Global content delivery and edge caching |
| TLS certificate management | Certificate provisioning and renewal |
Needs product confirmation: The specific infrastructure provider(s) in use; data residency options per workspace; the complete list of data regions for all services.
Section 2: AI model providers (tenant-configured)
AI model API calls are made to the provider configured by the tenant for each agent. ProvenanceOne routes the API call to the selected provider's endpoint. The following providers are confirmed in the platform's agent model:
| Provider | When data is sent | Status |
|---|---|---|
| Anthropic | When an agent is configured to use a Claude model | Confirmed |
| OpenAI | When an agent is configured to use a GPT model | Confirmed |
| Google (Vertex AI or direct API) | When an agent is configured to use a Gemini model | Confirmed |
| Azure OpenAI | When an agent is configured to use an Azure OpenAI model | Needs product confirmation (see note) |
Needs product confirmation: Whether Azure OpenAI is generally available to tenants or in private preview; what Azure region is used for Azure OpenAI API calls.
What data is sent to AI providers
When an agent step executes, the following data may be included in the API request to the configured AI provider:
- The agent's system prompt
- Workflow context at the time of execution
- Tool descriptions (the list of tools available to the agent)
- Tool call results (outputs from MCP tools the agent has already invoked)
- Agent persistent memory key-value contents (if memory is configured)
ProvenanceOne does not control what these providers do with this data. Each provider's enterprise data handling terms, data retention practices, model training opt-out policies, and sub-processor lists are governed by that provider's agreements.
Customers should review the following before enabling each provider:
- Anthropic's privacy policy and usage policies
- OpenAI's enterprise privacy and data handling
- Google Cloud's data processing terms
- Microsoft Azure's data privacy documentation
Needs product confirmation: Whether ProvenanceOne has zero data retention agreements or enterprise API agreements with any of these providers; whether data sent to AI providers is subject to any model training by default; whether a Data Processing Agreement between ProvenanceOne and each AI provider is in place.
Section 3: Billing
| Subprocessor | Purpose | Status |
|---|---|---|
| Stripe | Billing, payment processing, and customer billing portal | Confirmed (billing portal endpoint POST /billing/portal exists in the API) |
Stripe processes payment card data and billing information. Stripe's data processing practices are governed by Stripe's privacy policy and DPA. For information on Stripe's sub-processors, refer to Stripe's sub-processors list.
Needs product confirmation: Whether a separate DPA is in place between ProvenanceOne and Stripe covering customer data; what billing data is retained by Stripe and for how long.
Section 4: Unconfirmed subprocessor categories
The following categories of subprocessors are not confirmed from the codebase. Customers conducting GDPR Article 28 due diligence should request explicit confirmation before contracting.
| Subprocessor category | Status | Notes |
|---|---|---|
| Monitoring and observability | Needs product confirmation | Not confirmed; may include tools such as Datadog or New Relic |
| Error tracking | Needs product confirmation | Not confirmed |
| Email delivery | Needs product confirmation | Platform transactional email is confirmed (Section 1); additional providers not confirmed |
| CDN and frontend delivery | Confirmed | See Section 1 |
| Analytics and product telemetry | Needs product confirmation | Not confirmed; PostHog key present in documentation planning artefacts but not confirmed as deployed |
| Customer support tooling | Needs product confirmation | Not confirmed |
| Full GDPR subprocessor register | Needs product confirmation | A complete, published register has not been confirmed |
Section 5: GDPR, DPA, and data transfer
Needs product confirmation: Whether a GDPR Article 28 Data Processing Agreement (DPA) is available for customers to execute with ProvenanceOne.
Needs product confirmation: Whether Standard Contractual Clauses (SCCs) or equivalent transfer mechanisms are in place for EU-to-non-EU data transfers (including data processed by Anthropic, OpenAI, Google, and Stripe, all of which are US-based entities).
Needs product confirmation: The process by which ProvenanceOne notifies customers of changes to its subprocessor list; the notice period before a new subprocessor is added.
Needs product confirmation: Whether personal data processed through AI model API calls is covered by a controller-to-processor DPA between ProvenanceOne and the AI providers, or whether the tenant is required to enter into a direct agreement with each AI provider.
Recommendation for procurement teams: Request the following documents from ProvenanceOne before signing:
- The complete, current subprocessor list with entity names, countries of operation, and processing purpose
- A copy of ProvenanceOne's GDPR Article 28 DPA (or confirmation of its availability)
- Confirmation of transfer mechanisms (SCCs, adequacy decisions) for any EEA-to-third-country data flows
- ProvenanceOne's subprocessor change notification process and notice period
Auditability
ProvenanceOne does not surface subprocessor API call details in the workspace audit log. Audit events record agent step executions, tool calls, approval decisions, and system events — not the underlying HTTP requests to AI provider APIs.
Needs product confirmation: Whether there is any logging or visibility into which AI provider endpoints were called during a workflow run, and whether that information is accessible to customers.
Limitations and open questions
| Item | Status |
|---|---|
| Published subprocessor register | Needs product confirmation |
| GDPR Article 28 DPA availability | Needs product confirmation |
| SCCs for EEA-to-non-EU transfers | Needs product confirmation |
| Subprocessor change notification process | Needs product confirmation |
| Azure OpenAI GA status | Needs product confirmation |
| Zero-retention agreements with AI providers | Needs product confirmation |
| Monitoring/observability subprocessors | Needs product confirmation |
| Analytics/telemetry subprocessors | Needs product confirmation |
| ProvenanceOne ↔ AI provider DPA chain | Needs product confirmation |
FAQ
Does ProvenanceOne share my data with AI model providers?▾
When an agent step executes, the data required to generate a model response — system prompt, workflow context, tool descriptions, tool call results, and agent memory — is sent to the AI provider configured for that agent (Anthropic, OpenAI, Google, or Azure OpenAI). ProvenanceOne does not control those providers' data handling practices. Customers should review each provider's enterprise data processing documentation.
Is ProvenanceOne GDPR compliant?▾
ProvenanceOne has implemented specific GDPR controls including a data erasure endpoint (POST /workspace/members/{userId}/erase) and PersonID field erasure in audit events. Whether a GDPR Article 28 Data Processing Agreement is available, and whether the full subprocessor chain is covered, has not been confirmed. Request a DPA from ProvenanceOne before relying on GDPR compliance claims.
What infrastructure does ProvenanceOne run on?▾
ProvenanceOne runs on enterprise-grade cloud infrastructure. The platform uses managed services for serverless compute, database, object storage, workflow orchestration, identity management, secrets management, transactional email, and encryption key management. Data residency and specific provider details can be confirmed with the ProvenanceOne team.
Can I restrict which AI providers process my data?▾
AI model providers are configured per agent within your workspace. Workspace admins control which AI providers are registered under Settings → Model Integrations. Agents can only call providers that have been explicitly configured. You can restrict provider use by not registering providers you do not want used.
Is there a DPA available?▾
Whether a GDPR Article 28 Data Processing Agreement is available has not been confirmed. Request this directly from ProvenanceOne before signing any contract if you have GDPR obligations.
Does ProvenanceOne use my data to train AI models?▾
ProvenanceOne does not build or train AI models. Data sent to AI providers (Anthropic, OpenAI, Google, Azure OpenAI) is subject to those providers' data handling and model training policies. Review each provider's enterprise data handling documentation to understand their training opt-out options.
Related pages
- Compliance Controls — Security controls implemented and framework status
- SSO and Identity Management — Authentication and identity provider information
- Data Handling — Data storage, retention, and encryption
- RBAC and Permissions — Access control model
- Audit Log — What is logged and how audit integrity is maintained