This month, the rules of the game changed. In early August 2026, Europe switched on continent-wide requirements under the EU AI Act — including transparency rules that mean AI systems must identify themselves to the humans they talk to. India moved earlier: the February 2026 update to the IT Rules introduced labelling requirements for synthetically generated content. If your app or website has a chatbot, an AI voice, or AI-generated content, this now concerns you directly.
We are not lawyers, and this post is not legal advice — it is an engineering team explaining what these rules mean in practice, because we have already built this exact compliance layer once. Google Play forced us to, long before any government did. More on that below.
What the new rules actually ask for
Strip away the legal language and the 2026 transparency wave comes down to three practical demands:
- Tell people when they are talking to AI. A chatbot or AI voice agent that could be mistaken for a human must disclose that it is not one. A simple, visible "You are chatting with our AI assistant" line at the start of the conversation covers the core of it.
- Label AI-generated content. Under India’s updated IT Rules, synthetically generated media needs labelling, and platforms carry take-down obligations for unlabelled synthetic content. If your product publishes AI-written text, AI images or AI voice at scale, it needs a labelling strategy.
- Know your risk category. The EU AI Act treats AI differently by risk level. A support chatbot is a transparency question. AI that makes decisions about people — hiring, lending, medical triage — sits in the high-risk bucket with far heavier obligations. Most business apps only need the first bucket, but you should know which one you are in.
Who should pay attention: any business with users in the EU or UK, any Indian platform publishing AI-generated content, and — practically speaking — anyone shipping through Google Play or the App Store, because the stores have been enforcing their own AI disclosure rules ahead of the governments.
The app stores got there first — and we have the scars
Here is the part most compliance articles miss: for mobile apps, this is not new. Google Play rejected our own AI app well before these laws arrived — and the approval checklist looked remarkably similar to what regulators now ask: clear AI disclosure, content moderation, a user report-and-block flow, and honest description of what the AI does. We documented the whole journey in our rejection-to-approval story.
That experience taught us something useful: disclosure is an engineering task, not a paragraph of legal text. It touches your onboarding screens, your chat UI, your voice scripts, your content pipeline and your store listing. Bolting it on later is always more expensive than designing it in.
A practical 5-point self-check
First message of every conversation, visible, unambiguous. If a reasonable user could think they are talking to a person, you have a gap.
Voice is held to the same standard. Our own voice alerts open with an AI identification line in the language of the listener — it costs two seconds and closes the biggest risk.
Text, images, audio — if AI made it and users see it, mark it. India’s IT Rules make this explicit for synthetic media; a small "AI-generated" tag in the UI plus metadata is the pattern we implement.
A report and block flow is what got our app approved on Play, and it is the direction every regulation points. If AI can produce something a user objects to, the user needs a button.
If AI decides who gets a loan, a job interview or a diagnosis, stop and get proper legal advice — that is high-risk territory where an engineering blog is not enough.
What compliance actually costs
For a typical business app — a support chatbot, AI voice alerts, some AI-generated content — the disclosure layer is genuinely small: UI copy changes, a labelling tag in the content pipeline, a report flow if you lack one, and a review of your store listings. When we build AI features now, this layer is included by default, because shipping AI without it means the product cannot launch anyway. Retrofitting an existing app is usually days of work, not months — the expensive version is ignoring it until a store review or a regulator forces the conversation on their timeline instead of yours.
Penalties under the EU AI Act can reach into the millions of euros for serious violations — but honestly, for most small businesses the nearer risk is simpler: a rejected app update or a platform take-down. Those hurt on a Tuesday, not in a courtroom.