The rules that were "coming" have arrived
I've been saying for the best part of two years that the EU AI Act would eventually stop being a slide in someone's compliance deck and start being something you have to actually do something about. That day is here. The Verge has a good rundown of the transparency obligations that are now live — chatbots have to disclose they're AI, and AI-generated or manipulated content (images, audio, video, text in some cases) needs some form of label or disclosure.
If you build or sell anything in the content-automation space — and a fair few of the people reading this run tools not unlike my own Masher suite — this isn't background noise. It's a Tuesday afternoon job you need to do this month, not a "someday" item.
Who this actually catches
The Act's transparency provisions are broader than a lot of people assume. Broadly, two buckets matter to us:
Chatbots and conversational AI. If a user could reasonably think they're talking to a human and they're not, you need to tell them otherwise, clearly, before or during the interaction. Customer service bots, AI sales assistants, anything with a chat window bolted onto an LLM.
Synthetic and manipulated content. Deepfakes obviously, but also AI-generated video, images, and audio that could be mistaken for the real thing. Text is a greyer area — the Act leans harder on audiovisual content — but if your tool is generating "news-style" articles, voiceovers, or video content that looks like it was made by a person, you want to be thinking about disclosure now rather than waiting for a regulator to explain the nuance to you.
What this means if you're building the tools, not just using them
Here's the bit that catches people out: if you run a SaaS that generates content for other businesses — think Article2Video, VidMasher, anything that spits out video or audio at scale — the obligation doesn't stop with you. Your customers, who are putting that content in front of EU users, likely have disclosure duties too. And if your tool doesn't make it easy for them to comply, they'll either quietly stop using it or start asking you some pointed questions.
The smart move is to build the labelling in at the platform level rather than leaving it to each customer to remember. A few practical things I'm doing across my own tools:
- Bake a disclosure option into the output. A visible watermark, tag, or metadata flag that says "AI-generated" isn't hard to add and it gets you ahead of the requirement rather than scrambling to retrofit it.
- Give customers a compliance toggle, not a compliance lecture. Nobody wants a wall of legal text. A simple on/off setting that adds the required disclosure is far more likely to get used.
- Document what your tool does and doesn't disclose. When a customer's lawyer asks, you want a clear answer, not a shrug.
The chatbot disclosure is the easy win
Of everything in this round of rules, the "tell people they're talking to a bot" requirement is the cheapest to fix and the most embarrassing to get caught missing. It's one line of UI copy. If you've got a chatbot on your own site or embedded in a product, go check right now whether it says anywhere that it's AI. If it doesn't, fix it today — this is not a hill worth dying on.
Don't wait for enforcement to bite
The pattern with EU digital regulation is always the same: quiet start, patchy early enforcement, then a wave of fines that makes headlines and suddenly everyone's compliance officer is very interested in your product roadmap. GDPR taught us that. The sensible approach isn't to wait and see who gets made an example of — it's to spend a day now auditing where your tools touch EU users and where disclosure is missing.
This isn't about slowing down what we build. Content automation isn't going anywhere, and honestly, clear labelling is good practice regardless of who's mandating it — trust is the actual product we're selling, not just the content itself. Turning raw automation into something people trust is the real alchemy here. Getting the disclosure right is part of that gold, not a tax on it.
If you run any AI content tool that touches European users, spend an hour this week checking your chatbots and your output pipeline. It's cheaper than the alternative.
— Wayne