If you publish AI-generated speech and any of your listeners are in Europe, you now owe them one sentence. Article 50 of the EU AI Act became enforceable on 2 August 2026, and the duty to disclose synthetic audio sits with you, the person hitting publish, not with the company that made the voice tool.
That is the whole answer. Everything below is about where the line sits, because the rule is narrower than the panic suggests and the wording is easier than you would guess.
What changed on 2 August 2026
The European Commission announced the start of enforcement on 31 July 2026. Its own summary of what the transparency rules require: "chatbots and other interactive AI systems will have to tell users they are dealing with AI, not a human. Deepfakes (images, videos, or audio that have been edited or generated using AI) will have to be labelled. AI-generated or altered content will also have to carry machine-readable marks so it can be detected more easily."
Read that last sentence twice. Two separate duties are hiding in one paragraph, and they land on two different parties.
Are you a provider or a deployer?
The Act splits responsibility. Providers build and ship the AI system. Deployers use it. If you record a script, generate narration, and publish the file, you are a deployer.
| Duty | Who owes it | Source |
|---|---|---|
| Mark synthetic audio in a machine-readable, detectable format | Provider, meaning the voice tool vendor | Article 50(2) |
| Disclose that published audio was artificially generated or manipulated | Deployer, meaning you | Article 50(4) |
| Tell people they are talking to an AI, not a human | Provider of the interactive system | Article 50(1) |
Article 50(2) is a vendor engineering problem: watermarking, provenance metadata, detectability. You cannot fulfil it and you are not expected to. Article 50(4) is a publishing problem, and it is one line of copy.
One nuance worth knowing, straight from the text of Article 50(2) on artificialintelligenceact.eu: the marking duty does not apply where systems "perform an assistive function for standard editing or do not substantially alter the input data provided by the deployer or the semantics thereof." The regulation itself already draws a line between editing help and fabrication, which is exactly the line practitioners have been arguing about.
What actually triggers the disclosure
Podnews covered the rule for audio people on 3 August 2026 and quoted the cleanest practical read available, from Production Expert. Manipulation "doesn't mean de-noising or repairing audio (or using AI editing tools); but it does mean that clear disclosure is required if you're generating whole phrases that a voice never actually recorded."
So the test is fabrication, not assistance. Most of what creators call "AI in my workflow" is repair, and repair is untouched.
Cleaning up room tone on a recorded interview: no disclosure. Removing a cough, levelling loudness, trimming filler words with an AI editor: no disclosure. Generating a sentence your host never spoke, in your host's voice, to patch a section you could not re-record: disclose it. Narrating an entire audiobook with a synthetic voice: disclose it.
Do old episodes need retroactive labelling?
No. Podnews said it plainly on 3 August 2026: "Anything made before today does not need retroactive labelling."
So you can stop imagining a weekend spent rewriting old episode descriptions. Set your template today, apply it going forward, and leave the archive alone.
Where does the disclosure go for creative work?
This is the part almost nobody quotes, and it is the part that saves your cold open. Article 50(4) says that where the content "forms part of an evidently artistic, creative, satirical, fictional or analogous work or programme, the transparency obligations set out in this paragraph are limited to disclosure of the existence of such generated or manipulated content in an appropriate manner that does not hamper the display or enjoyment of the work."
In plain terms: a fiction podcast does not have to interrupt the story with a compliance announcement. A novel does not need a robot voice reading a legal notice before chapter one. Credits, show notes, and the product description are appropriate places. The regulation cares that the information exists and is findable, not that it is loud.
Non-fiction sits differently. A documentary or news show is closer to informing the public, so put the disclosure somewhere a listener meets it early.
Copy-paste disclosure lines
Write these once. Save them in your production template. Adapt the voice to match your show.
For an audiobook credit page or the end of the final chapter:
This audiobook is narrated using a synthetic voice. The text was written by [author name]; the narration was generated with AI voice software and reviewed by a human editor.
For a shorter version that fits a storefront description field:
Narrated with a digital voice. Written and edited by humans.
For podcast show notes, non-fiction:
Parts of this episode's narration were generated with AI voice software. Interview audio is unaltered apart from noise reduction and levelling.
For podcast show notes, fiction or scripted drama:
Character voices in this series are AI-generated. Scripts are written by [name].
For a video description on YouTube or elsewhere:
Voiceover in this video is AI-generated. Script, research and edit by [name].
Each of these does the same three things: it names what was generated, it separates that from what was recorded, and it keeps a human visibly in the loop. Put the line where a listener would naturally look for credits, not where it competes with your first ten seconds.
What YouTube and Spotify already ask you to declare
Platform rules are not the same as the regulation, and they are narrower in ways that surprise people.
YouTube's own disclosure guidance requires the altered-content label for realistic AI content that makes a real person appear to say or do something they did not, alters footage of a real event or place, or generates a realistic scene that never happened. You set it in YouTube Studio under Attributes, "AI use". On YouTube's explicit no-disclosure-needed list: "Cloning one's own voice to create voice overs or dubs", along with audio repair, upscaling and caption creation. YouTube also states plainly that "Disclosing AI content won't limit a video's audience or impact its eligibility to earn money."
Spotify for Authors handles audiobooks with a single toggle. Spotify's help documentation says "We accept audiobooks with digital voice narration on Spotify," and asks you to select "This audiobook uses digital voice narration" at upload. Spotify then adds a short sentence to your book's description so listeners know it uses digital narration. Worth knowing before you plan distribution: Spotify says it does not currently share audiobooks with digital voice narration to referral partners.
Ticking the toggle is the easiest possible evidence that you disclosed. Do it, and put your own line in the copy as well.
A ten-minute pass on your workflow
Nothing here requires a lawyer or a new tool. Write down which parts of your process fabricate speech and which parts only clean it. Then add your disclosure line to the template you already use: the show notes boilerplate, the credit page in your manuscript, the description field in your upload preset. Tick the platform toggle where one exists, and keep a note of which voice you used on which project. If anyone ever asks, "I generated the narration with a licensed synthetic voice and disclosed it in the credits" is a complete answer.
If you are setting up a show from scratch, the podcast voice kit includes the show notes and episode templates where this line belongs. For long-form narration, the audiobook production kit covers the credit page and chapter structure. Both assume you keep voice work on your own machine, which is the case we made in why privacy-first voice tools matter, and it makes provenance easier to answer honestly.
On the ethics of publishing a voice that is not the speaker's own, morphed voice ethics goes deeper than any regulation does. And if audiobooks are your channel, ACX audiobook compliance with AI voices covers a stricter gate than anything the EU imposes.
Vois runs generation locally under one flat monthly price, so disclosure changes nothing about how much you produce. Check the pricing page, then get back to the work.
The rule asks you to be honest about one thing in one sentence. Write the sentence, save it in your template, and never think about it again.
The Vois Team