AI Disclosure

Last updated: August 17, 2026 Status: Revised August 17, 2026 to describe Brin's hosted service. The revised sections are under counsel review. The limitations of AI output (Section 5) and the human-in-the-loop commitments (Section 6) are unchanged from the counsel-reviewed text.

Plain-language summary

Brin uses large language models (LLMs) and other AI services to power its most useful features: John (your Chief of Staff), the specialist AI workers, voice mode, planning, drafting, and knowledge base extraction. You should understand three things:

1. What models we use and who runs them. Brin calls models run by Anthropic, ElevenLabs, and OpenAI. These are not our models. They are run by third parties you should evaluate independently.
2. How your content reaches them. Brin sends AI requests on your behalf, from our systems, using our own provider credentials. We send only what a feature needs, not your whole workspace.
3. AI output is not perfect. LLMs can be wrong, outdated, or make things up. Always use your judgment before acting on anything Brin tells you.

This document explains each of these in more detail.

1. What AI models Brin uses

Brin is an application layer on top of third-party AI providers. We do not train or run any models ourselves.

We may add, remove, or swap providers. When we do, we will update this page and the sub-processor list at /sub-processors.

2. What data gets sent to AI providers

When you use an AI feature, Brin sends the AI provider:

• The prompt. This is the instruction Brin builds to get a useful response. It typically includes a system instruction, a short portion of your recent conversation, and relevant context from your workspace, such as your goals, top projects, recent messages, or the lead you are writing to.
• Your specific request. The task, question, URL, or voice input you just submitted.

Because Brin is a hosted service, these requests are sent from our systems using our own provider credentials, not from your browser and not with an API key of yours. You do not need a provider account, and the providers do not bill you for your use of Brin.

The provider processes the prompt, generates a response, and returns it. Brin then stores that response in your workspace as part of your history, so your conversations, drafts, and deliverables are there when you come back.

3. What is not sent to AI providers

The following are never sent to an AI provider unless they are explicitly part of a prompt you trigger:

• Your full task list, project list, lead list, or knowledge base
• Your entire conversation history with John
• Your login credentials
• Credentials for services you have connected, such as an email sending identity or a mailbox token
• Files or documents you have not explicitly shared with an AI feature

How your workspace data is stored and protected is described in Section 7 of the Privacy Policy.

4. How AI providers handle your data

Each AI provider has its own privacy policy and data handling commitments. You should read them. At the time this page was last updated:

• Anthropic (Claude): does not use API inputs or outputs to train its models by default. See Anthropic's Privacy Policy and API Usage Policies.
• ElevenLabs (voice): see ElevenLabs' Privacy Policy and Terms.
• OpenAI (Whisper): does not use API inputs or outputs to train its models by default. See OpenAI's Privacy Policy and Enterprise Privacy.

We do not use your content to train AI models, and we do not permit our providers to do so on our behalf. Because provider policies can change, always check a provider's current terms before relying on a specific data handling commitment. Brin is not responsible for how third-party providers handle data beyond the commitments we have contracted for.

5. Limitations of AI output

Large language models have known limitations. Brin cannot eliminate them. You should assume any AI output might be:

• Factually wrong — the model may confidently state something that is not true ("hallucination")
• Outdated — the model's training data has a cutoff; it may not know about recent events
• Incomplete — the model may skip over important points
• Biased — the model may reflect biases present in its training data
• Inconsistent — identical or similar prompts may produce different answers

Do not use Brin as a source of truth for:

• Legal, medical, financial, tax, investment, or professional advice
• Safety-critical decisions (medical, industrial, transportation)
• Identity verification or background checks about real people
• Anything that could cause irreversible harm if wrong

For these purposes, consult a qualified human professional.

6. Human-in-the-loop principle

Brin is designed as a copilot, not an autopilot. You are always the decision maker. The application:

• Shows you what John is about to do before taking action in most cases
• Lets you cancel or edit any AI-generated content before it is saved
• Routes work produced by AI workers into a review queue, so you approve a deliverable or send it back before it counts as done
• Requires you to review and approve every Lead Center email before it is sent, from your own address
• Does not automatically send emails, publish content, or execute financial transactions without your explicit instruction

If you ever find a feature that takes an action you did not approve, please report it to privacy@brinsolutions.com so we can fix it.

7. Using Brin without AI features

AI is central to how the hosted service works, so it cannot be switched off wholesale the way it could in a self-managed install. What you can control:

1. Voice is optional. If you do not use voice mode, no audio is sent to ElevenLabs or OpenAI.
2. AI features are triggered by you. Nothing is sent to a provider until you ask for something that needs it.
3. You can ask us to limit processing. If you need a narrower arrangement for your workspace, contact privacy@brinsolutions.com and we will tell you honestly what is and is not possible.

Choosing not to use an AI feature does not delete data you have already created.

8. Reporting problems with AI output

If an AI feature in Brin produces output that you believe is harmful, incorrect, or inappropriate:

• Report it to us: privacy@brinsolutions.com
• Report it to the provider: our AI providers each publish trust-and-safety reporting channels on their websites

We take quality and safety reports seriously and use them to improve prompts, guardrails, and documentation.

9. Changes to this page

When we add, remove, or swap AI providers, or change how Brin constructs prompts in a material way, we will update this page, update the "Last updated" date, and bump the LEGAL_POLICY_VERSION constant inside Brin so you are re-prompted to acknowledge the changes.

10. Contact

Questions about AI features or this disclosure:

• Email: privacy@brinsolutions.com