When AI Can Pass as Human, What to Disclose

A demonstration that circulated in late 2026 made the issue concrete: a full-duplex AI video system held a one-minute call, and roughly half the participants believed they were talking to a real person. Whatever you think of the specific claim, the direction is clear. AI output, text, voice, and now video, is crossing the line where a reader cannot tell. That puts a plain question in front of anyone who communicates: when do you have to disclose that AI was involved?

The Line Is About Deception, Not Assistance

Start with the principle, because the rules follow it. The problem is not that AI helped you write; it is a reader being deceived about who or what they are dealing with. Using AI to tighten your own email is assistance. Presenting an AI as a human counterpart, or passing off generated testimony as a real person’s, is deception. The disclosure obligation tracks that distinction.

The EU AI Act encodes a version of this in its transparency rules: people must be told when they are interacting with an AI system, and certain AI-generated or manipulated content must be marked as such. You can read the transparency provisions on artificialintelligenceact.eu. The underlying ethic is older than the law: do not let someone believe a machine is a person when it matters to them.

A Practical Rule For Writers

For the writing most people do, a simple rule covers the ordinary cases. If AI helped you say your own thing, your facts, your decision, your voice, you generally do not need to announce it, any more than you announce using a spell-checker or an editor. If AI is standing in for a human presence, or generating content presented as a real person’s experience, you must be clear it is AI.

Between those poles sit judgement calls. A support chat should say when a bot, not a person, is answering. A testimonial must be from a real customer, not a generated one. A mass message that implies personal attention it did not receive is edging toward deception even if every word is true. When unsure, ask whether a reasonable reader would feel misled to learn how it was made. If yes, disclose.

Rewriting Keeps You On The Right Side

This is where rewriting, as opposed to generating, is quietly reassuring. When you write the substance yourself and use AI only to improve the phrasing, the content is genuinely yours. There is no deception to disclose, because the facts, the judgement, and the position came from you. The AI changed the words, not the authorship.

Before, a generated draft you did not write:

Generate a heartfelt customer testimonial about how our product changed someone’s life.

After, your own words improved:

Rewrite this note from a real customer for clarity, with their permission. Keep their meaning and every fact exactly. Do not embellish or invent detail.

The second stays honest because the experience is real and the words only got clearer.

A Wrivio Context for honest rewriting could say:

Rewrite this in my voice for clarity and tone. Keep every fact and claim exactly as written. Do not add experiences, opinions, or details that are not in the original.

Press Ctrl+Shift+Space, paste the draft, and check the diff for anything the model added that you did not actually say.

Build It Into Your Norms

Rather than deciding case by case, set a standing norm: AI may polish your own words freely, and anything presented as a human’s words or presence is disclosed or not done. That norm is easy to follow and keeps you clear of both the law and the ethics. For how organisations are writing this down, see new AI disclosure laws and AI transparency and disclosure obligations.

Common Questions

Do I have to disclose that AI helped me write an email?

Generally no, if AI helped you say your own thing with your own facts and judgement. That is assistance, like using an editor or a spell-checker. Disclosure is about preventing deception, not about announcing every tool.

When is disclosure required?

When an AI stands in for a human presence or when content is presented as a real person’s words or experience. A support bot should say it is a bot, and a testimonial must be from a real customer, not generated.

What about the borderline cases?

Ask whether a reasonable reader would feel misled to learn how the content was made. If yes, disclose. A mass message implying personal attention it did not receive edges toward deception even when every word is true.

How does rewriting help me stay honest?

When you write the substance and use AI only to improve phrasing, the content is genuinely yours, so there is nothing deceptive to disclose. The AI changed the words, not the authorship, facts, or judgement.

Download Wrivio for Windows to improve your own words without letting a model invent experiences you never had.