How your data is handled when we build on it.

Lingows AI builds AI systems that run inside client environments and against client data. During an engagement we may receive sample documents, database access, API credentials, and production records. The sections below describe how that access is handled, how data moves to model providers, and what happens when something goes wrong.

Nothing here replaces the contract. Where a client agreement, data processing addendum, or security exhibit says something different, that document controls. Our draft data processing addendum and AI data policy carry the written terms.

We ask for the minimum data required to build and evaluate the system. In practice that means redacted or sampled records during assess and design, and scoped access to live systems only once the build requires it. Where a synthetic or masked dataset can carry the evaluation, we prefer it.

Client data stays in client-controlled systems wherever the architecture allows. When we need a working copy for development, it lives in the environment agreed for that engagement and is removed at the end of it.

Access is granted per engagement, to the named engineers on that engagement, and through your own identity and secret management wherever you have it. We ask clients to issue credentials that they can revoke without involving us. Shared logins are declined.

When an engagement ends, we ask you to revoke our access as a matter of course rather than relying on us to report that we no longer need it.

We are vendor neutral and route each task to the model that fits it. That means your data may be sent to one or more third-party model providers as part of normal operation. Which providers are involved is decided during design, documented in the system design document, and agreed before any production data flows.

If a workload cannot leave your perimeter, say so during assess. Open-weight models can be run inside your own network or a private cloud tenancy, and we will design to that constraint rather than around it. The tradeoff is usually capability and cost, and we will put numbers on it.

Does client data train models?

Our position is that client data must not be used to train third-party models, and we select provider tiers and account settings with that intent. The controlling detail is the provider agreement in force for your engagement.

Two retention questions matter and they have different answers. The first is what Lingows retains: engagement materials, evaluation sets, and any working copies of client data. The second is what a model provider retains, which is set by that provider's policy and the account tier in use, not by us.

For systems we operate after launch, logs and evaluation records are retained so that quality and drift can be measured over time. What those records contain is a design decision. Where the underlying data is sensitive, we log identifiers and outcomes rather than payloads.

Delivering an AI system involves third parties: model providers, cloud hosting, observability, and in some cases vector storage. Which ones apply depends on the architecture chosen for your build, and they are named in the design document before anything is deployed.

We will not add a subprocessor that touches client data mid-engagement without telling you first.

If we become aware of unauthorized access to client data in our possession, we notify the client contact for that engagement, stop the activity that caused it where we can, and preserve what is needed to understand the cause. We treat suspected incidents as incidents until they are ruled out.

A separate category worth naming: AI systems can fail by producing wrong output rather than by leaking anything. Systems we operate carry evaluation monitoring so that quality failures are caught as failing checks and reported to you, not discovered downstream.

Lingows AI does not hold SOC 2 Type I or Type II, ISO 27001, or any comparable certification. We do not claim HIPAA compliance, PCI DSS compliance, or FedRAMP authorization. We have not published penetration test results. If your procurement process requires any of these, tell us during assess so you can decide early rather than late.

Everything above is a description of how we work. If you need a control written into a contract, it belongs in the agreement, and we will negotiate it there.