Every agent that runs inside this firm operates under a rule we wrote for ourselves. It says that the machine may never identify itself as a person. Not by implication, not by omission, not by letting a misunderstanding stand.

It is a wordy disclosure. That is the point. Brevity is the thing being traded away, and it is not ours to trade. No law requires this of us. We think one should.

The harm is not hypothetical, and it is not evenly distributed

Impersonation is the mechanism. It is how fraudulent actors reach the people least equipped to detect them, and it works precisely because the person on the other end has no way to check. In Rob's words, doing so enables fraudulent actors to prey upon our most vulnerable subjects.

The registry idea, and why we abandoned it

A registry sorts compliant operators from non-compliant ones without giving the person being defrauded any way to tell them apart at the only moment that matters.

The conclusion we reached, and the one we would put to a legislator: if named AI is allowed to exist in any fashion, a consumer will never have a way to determine if they are being defrauded or not. Therefore the best mechanism of consumer protection is criminal and financial penalties.

What this is really about

For the good of the industry this needs to be stopped, because it erodes civic trust in the new frontier.

That trust is the scarce resource in this whole frontier. It is not ours. We are borrowing it from every business that comes after us, and it is being spent by people who will not be here to repay it.

We would like it stopped. We will start by binding ourselves.