AI Terms

What this covers

These terms apply to AI features Midcrowd builds for you as part of a custom project — agents, assistants, document and data workflows, search over your own content, and AI added to a system you already run.

The Midcrowd subscription platform does not currently put an AI feature in front of your website visitors or process your site content through a model. If that changes, this document changes with it and the change is dated.

AI output can be wrong

Language models produce plausible text, not verified fact. An answer can be confidently wrong, incomplete, out of date, or subtly different on a second run of the same input. No prompt, no retrieval system and no amount of testing removes that — it is a property of the technology, not a defect we have yet to fix.

You are responsible for reviewing output before it is acted on, published, sent to a customer or used to make a decision that matters. Where a mistake would be expensive or irreversible, the design should put a person in the loop, and we will say so when we scope the work.

AI output is not legal, medical, financial, tax, safety or other regulated professional advice, and must not be presented to anyone as if it were.

Third-party models

AI features are built on models operated by third parties. The providers used on your project are named in the proposal. Your prompts and the data you send are processed by those providers under their own terms and privacy notices, which you should read — they, not we, decide what happens inside the model.

Providers change their models, their prices, their rate limits and their terms, and they retire models. We may need to move your feature to a different model or provider as a result. Output can differ after such a move, and we will tell you before making one where we can.

Model availability is not guaranteed. An outage at a provider is an outage in your feature, and neither we nor you control it.

Usage and cost

AI providers charge for use. Those charges are yours, billed by the provider on accounts held in your name wherever that is possible. Usage is not fixed: a busier month costs more, and a feature that is called in a loop by mistake can cost a great deal more. We will design sensible limits where the work allows for them and tell you where the exposure is.

Your data and your prompts

You decide what goes into an AI system we build. Do not send personal data, customer records, credentials or confidential material into one unless the project was designed for it and you have the right to do so. If the system reads your documents or your database, everything it can read is potentially reachable through it — which is why access scoping is part of the build rather than an afterthought.

Systems that act on their own

An AI workflow can be built to do things — send a message, update a record, place an order, call another system. Anything it can do, it can do wrongly. For actions that spend money, contact a customer, or change data that is hard to restore, we recommend and will normally build an approval step, and we will tell you plainly when you ask for one without it.

What AI we build may not be used for

• Anything unlawful, or anything that infringes someone else's rights.
• Impersonating a real person or organisation, or presenting AI output as a human when a person would reasonably need to know.
• Generating deceptive content, scams, or material designed to manipulate or harass.
• Making decisions about a person's employment, credit, housing, insurance, health or legal position without human review.
• Anything the underlying provider's own usage policy prohibits.

How to contact us

Questions about this document can be sent to midcrowd.ai@gmail.com. We aim to respond within 5 business days.

Midcrowd is the name this service trades under. The registered legal entity and address are to be confirmed before these terms are final.