From now on, AI transparency is the law in the EU: 🙌 AI-generated content must carry a machine-readable mark. 🙌 Deepfakes and AI text on public interest topics must be visibly labelled. 🙌 When you chat with a bot or virtual agent, it must make clear it’s not a human. 🙌 When AI is actively analysing you, companies must tell you. We are building a system of trust and supporting companies with clear guidance. While you get the protection you deserve. This is how AI works for people in Europe 👉 https://lnkd.in/enSiCDqz * Updated to better clarify the marking and labelling obligations under the AI Act.
Does this ruling apply even to things like explanatory figures in research papers? Using AI to improve their style has become widespread in my field and having to label each of them (in the figure itself or caption, I suppose) is quite removed from reality. I would prefer to be treated like an adult.
That's a point European Commission but what about the other principles of the Responsible AI? In Responsible AI we still lack a shared taxonomy: disclosure of AI nature ≠ explainability of decisions ≠ transparency of training data and governance processes. Without clarifying what we actually mean by "transparency", we risk reducing a complex principle to a labelling obligation.
This is exactly the kind of framing that worries me. The real question isn’t “Was AI used?” The real question is “Is the content deceptive?” By reducing the discussion to AI disclosure, we’re shifting attention away from the quality, truthfulness, and accountability of the content itself. Ironically, this creates an environment where people can dismiss accurate arguments as “AI slop” without engaging with them. The origin becomes more important than the substance. Public institutions should be especially careful not to replace nuanced legal obligations with simplified social norms.
You’re not building a system of trust. You’re building a system that punishes honest disclosure and rewards anyone who simply leaves the label off. The best deceptions will not carry a warning, while unlabelled content will appear more trustworthy by default. That is not transparency. It is a perverse incentive.
ilteriş Öney
1w
It is good idea but how it will be enforced is a question. Like in gdpr websites that we are forced to accept conditions, similarly they may force us to accept AI conditions otherwise dont visit us, do not help.
European Commission Thanks for raising awareness. One point of feedback: the first statement may unintentionally suggest that all AI-generated content must be labelled. The AI Act is more nuanced and distinguishes between different use cases and obligations. A brief clarification might help prevent misunderstandings.
Ben Luong
1w
The telling thing is none of it even tries to save jobs. There's no mechanism anywhere in the Act that keeps a single person employed. We all still lose our jobs, just in a fully documented, risk-assessed, compliant way. Genuine question: under the rule you're announcing, shouldn't this post be labelled? The emoji listicle and the one-line paragraphs read like AI-generated copy. If transparency is the law from today, the announcement itself seems a fair place to start.
Marvin Doose
1w
What is a AI-generated content? Or should be asked, at what point is it considered a AI-generated content? 1. Someone write down all the ideas / concepts but just request AI to reformulate it clearly under a specific prompt without altering ideas / concepts. Is it AI-generated content? 2. AI is used to explore ideas but every points is deeply checked by human and then AI just used to prepare the final content which will be reviewed and validated by a human, word by word? Differences exist but applying it, it looks really complicate on a daily basis by « AI users » …
PolicyGate
1w
Worth being precise on "from now on," because the timelines split. The transparency duties here (Art. 50) apply from 2 Aug. But the high-risk obligations — recruitment, credit scoring, biometrics — were deferred to Dec 2027 under the Omnibus. The disclosure clock is live today; the record-keeping clock isn't. Both matter, and teams are quietly conflating them.
How do you want to differentiate an AI-generated content from the content generated by a human by means of AI? If I prompt that I want to write a text (let's say a blog) on a certain topic while lining out what it should contain, providing examples from my experience and my thoughts, and edit the AI-output after it's been produced, I don't want the result to be marked as AI-generated as the intellectual property is still with me. Most of scientific papers and research proposals are now written like I described. Seriously, you don't want them to be marked as AI-generated as opposed to the truly AI-generated content for two reasons: 1) you'd automatically strip the authors of their IP rights, 2) you'd further undermine public trust to research which is already severely jeopardised by social media and conspiracy theories.