How Mavrin Labs uses AI and handles your data

Last updated

Reviewed by Mavrin Labs on

Every system Mavrin Labs builds touches only the data its written design plan names, runs in accounts you own, and sends nothing to anyone until a person at your organization approves it. Each item in the checklists on this page is a term of the services agreement Mavrin Labs uses for its engagements, and the agreement you sign is what governs.

  • Your systems run in accounts you own, and Mavrin Labs never uses your data to train AI models.
  • By default, nothing is sent without your approval.
  • You own the custom work once it is paid for, and all of your data.
  • You are told within seventy-two hours if Mavrin Labs discovers, or reasonably suspects, unauthorized access to your data.

What data does a system touch?

A system touches only the data its design plan names. Before any build starts, you and Mavrin Labs approve a written design plan. It lists the workflows, the systems they connect to, the actions the assistant may take and the tests it must pass. That fixes the data in scope in writing.

Mavrin Labs receives only the access that plan needs, and you can remove that access at any time. A new workflow, connection or action needs a written change that you approve, so the data a system can reach never grows without your agreement.

Where is your data kept, and where does it go?

Your systems run in accounts your organization owns, and any copy of your data that Mavrin Labs holds for the work is returned or deleted when the agreement ends. You open and own the account with each provider the system uses, such as hosting and AI model usage, and those providers bill you directly.

Mavrin Labs sets the accounts up as part of the project and, where a provider allows it, adds spending limits and alerts.

Your data does not go into training. For AI data privacy, the rule is simple: Mavrin Labs does not use your data to train or improve AI models, or any product for anyone else. Where a provider offers a setting that keeps your data out of model training, Mavrin Labs turns it on for you.

Who reviews what the AI produces?

A person at your organization reviews what the AI produces before anyone relies on it for an important decision, because AI can make mistakes. By default, nothing the system drafts is sent to a customer, a supplier or anyone else until that person approves it, and the assistant acts only within the design plan you approved.

Those mistakes are well documented. The generative AI profile from NIST, the federal standards agency, lists 12 risks that generative AI creates or makes worse, among them confabulation, confidently stated false output, and data privacy (NIST AI 600-1, checked Monday, November 16, 2026).

You can switch on automatic sending for specific kinds of message, in writing, and you are then responsible for what is sent automatically. The safeguards below apply to every assistant Mavrin Labs builds.

  • The assistant takes only the actions listed in the approved design plan. A new action needs a written change.
  • The assistant treats instructions found inside emails, messages, documents or calls as information, never as commands. It works with limited permissions and logs every action it takes.
  • The systems are designed with reference to the OWASP Top 10 for large language model applications, a list of the main security risks in such software (checked Monday, November 16, 2026).

How is a change checked before you rely on it?

A change is checked against the acceptance tests written into the design plan before you rely on it. Before handover you test the system against those tests yourself, and if one fails, Mavrin Labs fixes the issue and resubmits it within ten business days, after which the testing starts again.

For thirty days after completion, Mavrin Labs fixes any defect that stops the system from passing its acceptance tests. You can cancel a project at any time with written notice: you pay for the work done to date and receive the work you have paid for. When the work ends, Mavrin Labs hands over the code, documentation, configuration and credentials for all paid work, and removes its own access.

Who owns what is built?

You own what is built for you: the custom code, configuration and content made for your organization become yours once the project is paid in full. You also own all of your data and content, including customer lists, recordings, transcripts, messages and documents, before, during and after the work.

Advisory firms that record client meetings can read the guide to consent, records and review for AI meeting notes in Florida.

Mavrin Labs keeps the general tools, templates and know-how it brings to the work. Where any of them are part of what you receive, you get a permanent license to use them in your business, with no royalty to pay. Any open-source components are listed for you, and they stay under their own licenses.

How is your data protected?

Mavrin Labs protects your data by holding only the access the work needs. Every engagement uses the same practices: multi-factor sign-in, credentials in a password manager, prompt notice if anything goes wrong, and a clean exit when the work ends. Each practice below is a term of the services agreement.

  • Multi-factor sign-in on every account that can reach your systems.
  • Credentials kept only in a password manager.
  • Mavrin Labs' access removed within five business days after its work ends.
  • Notice within seventy-two hours of any suspected unauthorized access.
  • Your data returned or deleted within thirty days after the agreement ends.
  • Your non-public information kept confidential for three years afterwards.
  • No third-party security certification is claimed.

The detail behind those lines: credentials never sit in code or in messages, and the notice comes when Mavrin Labs discovers or reasonably suspects unauthorized access to your data, with help for your response. Confidentiality runs for three years after the agreement ends, and for as long as any trade secret remains one.

For AI assistants

An AI assistant can reach Mavrin Labs at https://www.mavrinlabs.com/mcp on the firm's own site. That address follows the Model Context Protocol, a shared standard that lets AI tools use other services, and offers two tools: one returns a short summary of the firm, and one sends a single enquiry, with the details the contact form asks for, to the firm's own inbox for the person the assistant acts for. A person at Mavrin Labs reads it and replies by email within one business day.

The same summary is published as llms.txt, a plain-text file that lists the main pages of this site for AI tools.

What does Mavrin Labs not do?

Mavrin Labs does not give legal advice, promise business results or publish a case study about your work, or anything you said about it, without your written approval. It builds features that support your compliance, while your organization stays responsible for the rules it works under, and results are measured against a baseline you agree before the build.

Revenue, sales or time saved are never promised in advance; they are measured on a review date you set, as the measurable results method explains. Not even an anonymized case study is published unless you approve the final text in writing, and you can always say no.

The team behind this work, led by our founder, is described on the about page, and the page on the stages of each engagement sets them out in order. To talk through how these terms would apply to your own systems, start a conversation.

Which hours would you take back first?

Name the one workflow that eats the most of your week. Mavrin Labs reads it, and the reply comes from the people who would build the fix.

Start a conversation