AI Disclosure

— 01 / OUR POSITION ON AI

Where we stand.

AI systems are powerful, useful, and imperfect. We build them because they solve real problems for real businesses — answering calls, following up on leads, handling repetitive back-office work at a scale humans can't sustain alone. We also believe transparency about how these systems work is a baseline requirement, not a nice-to-have.

This page is our attempt to be clear about where AI shows up in our business — on our own site, in the services we deliver, and in how we handle the data involved.

— 02 / AI ON THIS WEBSITE

AI on this website.

Some content on this website — including illustrative agent transcripts, marketing copy drafts, and technical explanations — was drafted or refined with the assistance of AI tools. All published content has been reviewed by a human before appearing on the site.

Metrics, statistics, and product claims shown on this site (call volumes, response times, integration counts, testimonial figures) may be illustrative or aspirational unless explicitly marked otherwise. Real customer data will be used only with the customer's consent, attributed clearly.

If this site includes a chat widget in the future, users will be told they are interacting with an AI, will have the option to request a human, and will not be subject to automated decisions producing legal or similarly significant effects.

— 03 / AI IN OUR SERVICES

AI in the agents we build.

The core of our business is designing, building, and deploying AI agents for our clients. When a client deploys one of our agents to their customers or end users, that client is responsible for:

• · Disclosing to their end users that they are interacting with an AI
• · Providing clear escalation paths to a human when appropriate
• · Complying with any AI-specific disclosure laws in their jurisdiction (including but not limited to California SB 1001, Colorado AI Act, EU AI Act, and equivalents)

We include recommended disclosure language in our onboarding documentation and help clients configure escalation paths. Ultimate responsibility for compliance rests with the client, who is the data controller and the party with the direct end-user relationship.

— 04 / MODEL PROVIDERS

Who provides the models.

We do not train foundation models ourselves. Our agents are built on top of models from third-party providers, which may include:

• · OpenAI (GPT family)
• · Anthropic (Claude family)
• · Google (Gemini family)
• · Open-source models (Llama, Mistral, and others) where appropriate

The specific provider used for a given client engagement depends on the use case, the client's preferences, and the client's own account credentials. Clients own and control the accounts through which their agents access these models — we do not proxy model calls through Miaste Stack infrastructure.

Each provider has its own terms, data-handling policies, and privacy practices. We review these before recommending providers to clients and include the relevant terms in engagement documentation.

— 05 / DATA AND TRAINING

Your data, model training, and what happens to inputs.

We do not use client data to train foundation models. Where our agents send inputs to third-party model providers:

• · We select provider configurations that disable use of inputs for provider-side model training
• · We select data-retention settings appropriate to the client's requirements
• · We do not retain conversation transcripts beyond the periods documented in the applicable Statement of Work

For our own internal improvements — refining prompts, tuning agent behavior, improving methodologies — we may use anonymized, aggregated performance metrics with all identifying information removed. We do not use identifiable client or end-user content for these purposes.

— 06 / HUMAN OVERSIGHT

Human oversight.

Every agent we deploy is configured with human escalation paths. No agent should be operating in a fully autonomous mode for use cases where a wrong output could cause meaningful harm.

Recommended and standard configurations include:

• · Escalation to a human on explicit user request
• · Escalation on detection of urgency, distress, or complexity outside the agent's scope
• · Human review of agent transcripts on a sampled basis during the first 30 to 60 days of deployment (part of our tuning phase)
• · Ongoing monitoring dashboards visible to the client

We advise clients against deploying our agents in fully automated pipelines where the output directly affects a person's legal rights, financial standing, healthcare, or safety, without a human in the loop. We reserve the right to decline engagements where this line would be crossed.

— 07 / CONTACT

How to reach us.

Questions about this AI Disclosure, or concerns about the behavior of an agent we've deployed, should be directed to:

privacy@miastestack.com

For behavior of a specific deployed agent, please also copy the client operating the agent — they are the direct data controller and the first line of response.