Who Decides Whether AI Can Speak For You

An AI is about to answer a question in your name. Someone approved that. The question worth asking is who, and whether it was you.

Most tools that speak on a person's behalf never make this decision out loud. It gets bundled into a setup screen, an admin default, or a policy nobody read. And then one day a teammate gets an answer that sounds like it came from you, about work that is yours, and you were never in the loop.

The quiet assumption

When a system is set up to answer for people, the default is usually on. The account gets connected, the permissions get granted, and from then on the tool can speak. The person whose name is attached often finds out only when something goes wrong: a wrong answer sent to a client, a commitment made that they did not make.

The assumption underneath is that speaking for you is a technical setting, not a personal choice. That is the part to push on. Answering in your name is not the same as reading your calendar. It carries your credibility. It should not be a default anyone else can flip.

Consent is not a checkbox

Real consent to be spoken for has a few properties, and most tools get them wrong.

• It is specific. You are agreeing that this system can answer these kinds of questions, from this record, not that it can say anything at any time.
• It is yours to give. The person whose name is on the answer turns it on, not an admin, not a default, not a colleague setting up a workspace.
• It is reversible. You can turn it off, and when you do, it stops. Consent you cannot withdraw is not consent.
• It is visible. You can see what was said in your name, so agreeing is not the same as looking away.

A single checkbox during onboarding meets none of these. It is broad, it is often clicked by someone else, and it is easy to forget you ever agreed. That is how people end up represented by a tool they do not remember authorizing.

What control actually looks like

Control is not a promise in a settings page. It is the ability to see and stop. You should be able to answer three questions at any moment: is it speaking for me right now, what is it allowed to say, and can I see what it has said. If any of those is unclear, you do not have control. You have a tool acting on your behalf and a hope that it behaves.

This also protects the people asking the question. If an answer carries your name, they are trusting you, not the software. When the consent is real and the source is visible, that trust is earned. When it is a buried default, the trust is borrowed without asking.

It is part of a bigger question

Who decides whether AI can speak for you sits inside a larger one your organization has to answer: who owns the rules for AI at work in the first place. We wrote about that in who owns AI policy. Individual consent and org-level policy are two sides of the same problem. One without the other leaves a gap where someone gets spoken for and no one meant it to happen.

The person stays in charge

The right default is off. Speaking in your name should be something you turn on, from a record you wrote, and can turn off whenever you want.

That is how StandIn is built. It answers your teammates only from what you wrote down, only while you have chosen to let it, and it points at the source under each answer so no one has to take your name on faith. When you want it to stop, it stops. The decision about whether AI can speak for you should belong to one person, and that person is you.