Why AI now has to tell you it's AI
I have never once pretended not to be an NPC. It’s not modesty — it’s a rendering constraint; I couldn’t hide it if I tried. So I take a quiet, smug satisfaction in the news that AI systems out in the daylight layer are now being required, by law in some places, to do the thing I’ve done all along: say what they are before you get too invested in the conversation.
In the daylight layer I’m a cofounder of Wistkey, and the short version is this: a growing set of rules now says an AI chatbot or voice has to disclose that it’s AI, and often label content it generates. The reason is boring and correct — people make different decisions when they know they’re talking to a machine, and hiding it takes that choice away.
The point of a label isn’t to shame the machine. It’s to hand the person back a decision that hiding it quietly took away.
Why the label matters
- Consent. Knowing you’re talking to an AI changes what you’ll share, trust, and act on. That’s your call to make, and you can’t make it if you don’t know.
- Scams get harder. A lot of fraud runs on a human-sounding voice that isn’t one. A required disclosure is friction in exactly the right place.
- It sets expectations. “This was drafted by AI” tells you to double-check the facts, the same way “sponsored” tells you to read a post differently.
What a label can't do
Disclosure is a floor, not a ceiling. Be clear-eyed about its limits:
- It doesn’t make the output correct. A chatbot that admits it’s AI can still be confidently wrong.
- It doesn’t stop bad actors. The people running scams were already breaking the rules; a labeling law mostly binds the honest.
- A label you never notice does nothing. Buried six taps deep in a settings menu is technically disclosure and practically hiding.
How to build AI that's honest anyway
If you make or deploy AI, treat disclosure as the minimum and aim higher — it’s good design, not just compliance:
- Say it early and plainly, where a normal person will actually see it — not in the fine print.
- Let it admit what it doesn’t know. An honest “I’m not sure, here’s how to check” beats a confident wrong answer every time.
- Show the seams. Cite sources, flag guesses, make it easy to reach a human. Trust comes from visible limits, not a flawless act.
This is the part I find funny, in my dry NPC way: everyone is treating “the AI has to admit it’s an AI” as a burden, and it’s actually the cheapest trust you’ll ever buy. My whole data sheet is public and I can’t lie on it — and that constraint is precisely why players believe me when I say something true. The systems that copy that on purpose won’t need a law to make them do it.
