--- title: "Security and AI data use | Momentic" description: "What Momentic sends to AI models, how long it keeps that data, who can read it, and what an Enterprise agreement changes. SOC 2 Type 2." canonical: "https://momentic.ai/security" last-updated: "2026-09-01T18:01:53-07:00" --- # Security and AI data use Momentic runs an AI agent against your application, so some of your test data leaves your environment. We send a model your page content and screenshots, and we keep the traces. Our engineers read a trace to debug a run when you ask. An Enterprise agreement changes what we retain and what we train on. - **SOC 2 Type 2:** Certified. Ask us and we send the report. - **Penetration tests:** A third party tests us every year, and the reports are in the Trust Center. - **Sub-processors:** Every sub-processor is published, including the AI providers. - **SSO, roles and audit logs:** SAML and OIDC on Enterprise. ## Certifications Momentic is SOC 2 Type 2 certified. The Trust Center holds the report, the security practices and the annual penetration test reports. Links: - [Trust Center](https://trust.momentic.ai/) - [Sub-processors and security docs](/docs/account/security) - [Privacy policy](/legal/privacy) - [Terms of service](/legal/tos) ## What Momentic sends to a model A request to a model can include the step instruction, page context, screenshots, an accessibility or HTML snapshot of the page, step traces, and the model output itself. An uploaded run can include video, DOM, network and console traces, screenshots, auto-heal traces and the agent's reasoning. Momentic reaches only the applications, pages and artifacts that your test configures. It does not crawl your systems and it does not ingest your document repositories. Tests are not designed to process personal data. ## Model providers Momentic uses several hosted model providers: Anthropic, OpenAI, Google Vertex, xAI, Mistral, OpenRouter, AWS Bedrock and Microsoft Azure. One provider outage does not stop your suite. Every model provider, and Langfuse, the observability provider that stores AI traces, is listed in the public sub-processor list. ## Infrastructure and encryption Momentic runs on Google Cloud Platform. Data is encrypted in transit and at rest. The public security docs list every sub-processor, with the purpose of each one. The list includes the AI providers, the hosting provider, and the tools we use to run the company. Links: - [Sub-processor list](/docs/account/security) ## Retention | Data | Retention | | --- | --- | | Application logs from the CLI (OTEL) | 14 days | | AI traces, in our observability provider Langfuse | Kept by default | | Run artifacts in the dashboard | See your plan's results retention | Momentic engineers read this data to debug a run when you ask us to. Internal access to the logs and to the traces is gated by SSO and role-based access control, and is limited to staff with a business need. An Enterprise agreement can add a zero-data-retention policy. ## Training Data from self-serve accounts can be used to improve Momentic's own agents. Enterprise customers can opt out of that use entirely, or agree a custom fine-tuning arrangement. ## Access control API keys are issued per user, per CI pipeline and per environment, and the value is shown once. SAML and OIDC single sign-on are available on Enterprise, and only an admin can manage the connection. Roles control workspace access, billing and membership. Audit logs cover administrative actions. ## Code execution in a test A JavaScript step runs in a Node sandbox or inside the page. Execution has a timeout and returns a structured result. In cloud runs, the evaluation happens on a separate endpoint that authenticates the calling organization. The step deliberately exposes real capability, such as an HTTP client, Postgres and GitHub helpers, because tests need it. Treat a JavaScript step as code you own and review. ## Prompt injection The agent never receives free-form user text. It receives structured page data, and it must answer through a strict tool schema. The prompt requires the agent to act on elements that exist in the current page, and to fail rather than invent an element or a value. A deterministic runner executes the result. Momentic runs no prompt firewall, because many customers test sensitive sites and a firewall would read that content. Momentic treats page content, traces and prior run memory as untrusted input. ## How to run Momentic safely - Point tests at a staging or test environment. - Use faked or cleaned data. - Keep production personal data, confidential documents and unnecessary secrets out of test flows. - Issue a separate API key per CI pipeline and per environment, and rotate it. ## Your tests stay in your repository A Momentic test is a YAML file in your own repository. You review it in a pull request, and npx momentic run executes it on your machine and in your CI. Your Git history is the audit trail. ## Questions ### Does Momentic train on my test data? On self-serve plans, yes. Data from those accounts can be used to improve Momentic's own agents. Enterprise customers can opt out of that use entirely, or agree a custom fine-tuning arrangement. ### Which AI providers does Momentic send my data to? Anthropic, OpenAI, Google Vertex, xAI, Mistral, OpenRouter, AWS Bedrock and Microsoft Azure. AI traces are stored in Langfuse. Momentic uses several providers so that a single provider outage does not stop your suite. ### What does Momentic send to a model? The step instruction, page context, screenshots, an accessibility or HTML snapshot of the page, step traces, and the model output itself. ### How long does Momentic keep my data? Application logs from the CLI are kept for 14 days. AI traces are kept by default in our observability provider. Run artifacts in the dashboard follow your plan's results retention. ### Can I get zero data retention? Yes, under an Enterprise agreement. That agreement can add a zero-data-retention policy. ### Is Momentic SOC 2 compliant? Yes, SOC 2 Type 2. Ask us for the report, or read the security practices and the annual penetration test reports in the Trust Center. ### Does Momentic need access to my production environment? No. Point tests at a staging or test environment, and use faked or cleaned data. Momentic reaches only the applications, pages and artifacts that your test configures. ### Who at Momentic can read my run data? Momentic engineers read it to debug a run when you ask us to. Internal access is gated by SSO and role-based access control, and is limited to staff with a business need. ### Is my data encrypted? Yes, in transit and at rest. Momentic runs on Google Cloud Platform. ### Can I take my tests with me? Yes. A Momentic test is a YAML file in your own repository. You review it in a pull request, and npx momentic run executes it on your machine and in your CI. Your Git history is the audit trail. ## Need the paperwork? Enterprise agreements cover zero data retention, training opt-out, SSO and audit logs. Talk to us and we will send the SOC 2 report. For a security question or a vulnerability report, write to support@momentic.ai. [Contact sales](/sales)