AI Disclosure

The Operator is an artificial intelligence system. We never present it as a person, and we built the product so you never have to wonder what it is allowed to do. This page states plainly what it is, how human oversight works, and where the limits sit.

What the Operator is

The Operator is an AI agent that runs on infrastructure dedicated to your business, connected to the systems you already use. It reads, drafts, coordinates, tracks, and executes the working procedures we configure with you. Its reasoning is powered by frontier language models; its authority is governed by a permission system we author with you, in writing, before it acts.

Human review is a setting you control

How much of your Operator's work passes through a person before it lands is an authored choice, made per action class during configuration. Many clients start with every outbound piece routed to a reviewer and widen autonomy as trust builds. Others authorize routine internal work to run unattended from day one. The system imposes no default in either direction: an action class you have not configured is refused, not guessed at. Whatever posture you author, every action is recorded in an audit trail you can read in your portal.

What it never does

Some limits are not settings. The Operator never sends communications impersonating you or your staff; it works under its own identity. It never gives itself more authority; only you and we can change what it is allowed to do, and every change is logged. It never performs licensed professional work: in a law firm it coordinates and drafts for attorney review but never advises or signs; the same boundary applies to every regulated profession we serve. And an integrity gate blocks specific fabrication classes before anything can be sent: citations it cannot verify, identifiers like account or case numbers without a traced source, and known fabrication tells in drafted text.

Where the AI runs, and your data

Language model inference runs through Anthropic's commercial API. Content passes through transiently when your Operator performs a task, and under Anthropic's commercial terms it is not used to train models. Your business data stays in your systems; the Operator's own working memory and audit record live on your dedicated machine. The full data-handling picture, including sub-processors, is on our security page and in your data processing addendum.

When AI is the wrong answer

Part of our job is telling you when it is not the right tool. Some problems are solved with a better process, a clearer role, or a simpler workflow, and we will say so rather than sell you an Operator you do not need. If you have questions about how AI is used in your engagement, write to team@smd.services.