待翻譯：EU AI Act Content Labelling: Fake Around and Find Out

AI 服務暫時不可用，以下為來源正文，待恢復後補全翻譯。

In this week's Thursday Thinkpiece we take a look at the recent EU AI Act and how it's changing the way the internet and businesses market themselves in this ever changing climate. There’s a place where creativity meets corporate interests. A place where selling a product or vertical slice of yourself is the goal.The place I speak of is a little place we call Social Media. On Social Media, you can be anything you want to be. You can turn yourself into an expert on any topic or industry. This is a place where you can run a whole business on vibes and journeys. A place where realism isn’t the name of the game but where a curated version of yourself can exist separate to you. You can deflect criticism quite easily but this is usually by just ignoring the “haters” who are just jealous of you of course. Money isn’t always the goal but AI has helped to turn everyone into a writer, artist or music megastar with nothing but a pocket full of dreams and a prompt full of hope. I can say I’ve leant on using Gen-AI to help me structure my thoughts and even to try and make something fit a more rigid structure. It’s not because I can’t write, it's more because it’s a skill I haven’t been able to practice properly for a while. It failed because that probably didn’t help to really bring forward what I was trying to achieve. It doesn’t know me, it just “knows” what I think I should be. It’s kind of the main reason why these think pieces even exist. As someone who loves to write, edit video, draw and play music in an analogue and sometimes digital fashion. I don’t want to take away anyone’s ability to express themselves. I can love Bob Dylan before he went electric and after. It doesn’t mean I’m amazing at any of these things but whatever I put on the page comes from my unique lens of the world. I then sprinkled in some inspiration from what has gone before. That’s it. However, the future of Gen AI trying to replace human ingenuity isn’t something far away on the horizon of civilization anymore. It’s here. But I digress. In its most basic form, Gen-AI. Well it’s just a tool. On Sunday August 2nd 2026, The EU AI Act started officially applying but it actually came into force on August 1st 2024 for those taking notes. It shifted what was thought to be a dream for those who didn’t necessarily fit the archetype of the tortured artist. To try and larp like one even if deep down they truly are. It’s a shortcut which now comes with up to a 15 million Euro price tag for those who don’t disclose it clearly, or 3% worldwide annual turnover whichever hurts them more. This is where FAFO stops being a bit of internet vulgarity and starts meaning Fake Around And Find Out. 15 million euros could be thought of as a rounding error to a big tech company like Meta. 3% of global turnover well that’s gonna sting Zuckerburg and co more than the loss made from the Metaverse. What is Generative AI? Generative AI is the synthetic output of large language models that have been trained on almost the whole of human histories’ art, music and literature to produce something which can only be described as an uncanny valley of humanity. This output can be in illustrated flyers for businesses, funny video shorts featuring animated pieces of fruit cheating on each other on YouTube and TikTok or as banal as social media posts on LinkedIn and X. It ticks all the boxes of what went before it and in many cases is probably the best way of doing things for efficiency and speed for individuals or small teams. The application of this technology is seemingly infinite and is not going away any time soon. What began as a way to save some time writing an email has now become the way to fully write, voice and produce a song with near studio level production. I’m not here to say it’s bad or that people should be penalised for using it as that is very Pre-ChatGPT thinking. The bottle this genie lives in was smashed to pieces many years ago. It’s not going back in. The wishes however now come with a bit of a catch. What is the EU AI Act? The EU AI Act first hit the headlines in 2023 (which I’ll come back to shortly) it seemed broad and a bit of a nothing burger way back when. This is how all legislation is initially drafted. Back in November 2022, when many of us early adopters signed up to the free OpenAI Playground I played around with it in its most basic form before ChatGPT Plus became a thing. Plenty of “at capacity” messages which I’m sure a lot of you remember and it was very basic and fiddly to set up. It was pay as you go and I was very careful in staying within very small limits. Then in February 2023 when ChatGPT Plus launched I was lucky to already have an account and credit card attached so I just bit the bullet to have the subscription to have a play around and get around the self imposed and technical limits. It was fun. Honestly, I mainly used it for shopping lists and asking stupid questions. I tried more ambitious things but it wasn’t really there yet. It wasn’t much more than that but organising my life as the goal here. It was very slow so it didn’t have the same kind of impact and I was deep into coming off a few years of learning how to use programming languages like Powershell and Python. At this point coding tasks were a lot easier to just do by hand and I had already caught the programming bug hard so hobbled things together from online courses and copious amounts of Stack Exchange posts. Those were the dark days indeed or maybe most fulfilling depending on how you look at it. Many people didn’t even know what Generative AI “content” was at that point. It wasn’t really usable or a thing until much deeper into 2023. This was mainly as it hadn’t really been defined fully yet and the edges were still pretty rough. It was years away in our minds. We were in the early days of chatbots and ChatGPT coming to take over the academic fields first and then coding. Governments around the world weren’t massively interested in trying to keep up with the technological changes speeding by but in the EU this was a different story. The EU AI Act covered a lot which I won’t go into but deepfakes were a major concern back then. Celebrities were worried their likeness would be used at best, to promote the wrong brand and at worst into more nefarious adult avenues. The broad outline was there to also protect the average citizen and state official from suffering a similar or worse fate. What actually changed on August 2nd 2026? Fast forward to August 2nd 2026 and this is now EU Law specifically Article 50 of this long sprawling document is the key change. Businesses have to become compliant by December 2nd 2026. This timeline and window is shocking but exists for a reason to try and allow those who may have started their business in the beforetimes a chance to align themselves closer to the new guidelines. This includes the marking and detection obligation of this content that they now face. There are some exceptions for things like satire especially if disclosure ruins the joke. Thanks Europe, I definitely owe you a stein of beer for that one and so do the Lego AI bros from Iran. This doesn’t just affect people like me in the UK though. We're bound by EU law of course even if we’re not officially a part of them any more. That does kind of make me sad but I don’t want to get into politics as that in itself would take us way off track. The main point though is this will also apply, just the same to other companies from around the world that trade here either physically or digitally. It mainly hits those in the marketing game and small to medium businesses who may not even really know about it. Again, this was still brought into law to in essence to fight the industry of deepfake technology which now is almost imperceptible from reality. We could have a whole other debate on the “someone please think of the children” when it comes to laws being brought in around technology so let’s put that to one side here even though I think it’s important. The main change is that these items now have to be labelled if they are to be used, especially if they are going to be used to inform, sell a product or try to educate. It’s probably for the best really. This is where things get a bit messy and of course there have been no actual test cases yet so it’s interesting to ponder on. It’s a very human trait. For those who want to read the actual letter of the law you can go here. Luckily the EU has been kind enough, if you could call it, to give us the actual labels to use and to allow some exceptions. Think of them as Food Warning labels and they definitely are going to make some people think a bit differently on the content they consume. I’ve pasted this below from the guidance from which makes it a bit easier to digest. Not all AI-generated or manipulated content needs to be labelled. The disclosure requirement under the AI Act covers only AI-generated or manipulated content. The icons support compliance with Article 50(4) of the AI Act, which requires deployers of AI systems to disclose: Deep fakes - AI-generated or manipulated image, audio, or video content that resembles existing persons, objects, places, entities or events and would falsely appear to a person to be authentic or truthful. AI-generated or manipulated text published to inform the public on matters of public interest, which did not undergo human review or editorial control and where editorial responsibility was not assumed by any legal or natural person. Link to Labelling and EU Gen AI Icons Making Gen-AI Iconic The icons are the thing that is really the major change here. I’ve already seen on LinkedIn content with the AI-Generated logo on videos this week. They now have this new watermark where only last week they didn’t. It was obvious they weren’t real to begin with but now they have to actually slap it on there to cover their arse for compliance purposes. Now many will say good riddance to the AI-Slop but it’s not as black and white as that and really topics like this never really are. I won’t even go into the Seems Like AI Slop button on LinkedIn as that’s a whole other piece coming later down the road. Let’s do a quick experiment for fun. Here’s an image that I made using Google Nano Banana It's a camera on a desk and I mean it did a really good job at making that camera on a desk. If I threw this up and didn’t say it was Gen-AI most would just think it was a stock image or maybe something from the Creative Commons free to use image library. However, these images contain watermarks that would give incorrect metadata so I can just use some simple python to strip the Exif Metadata and if I was really cheeky could even inject my own and voila this would seem to be the real deal. The funniest part is I just asked Google Gemini how to do this as well so the whole idea of the metadata being the way to truly catch people out is probably not going to work. Plus it’s kind of still baked into a tool that many use every day. In short, I made the image using Generative AI and then stripped out the important watermark data it’s now supposed to have to say it was created using that same Generative AI. The real test and way to become compliant is that now companies will have to either disclose it via the tags we just removed (C2PA Tags for technical resilience) and the EU Mandated Labels with a short paragraph on which AI tool this was created from if used for a blog or video for example. This is of course voluntary disclosure but if they want to challenge this and get called out on it then they in this case they would have to show their human work on how this picture was composed to start with. Maybe they need to document and take copious notes of the people they hired to take that photo. Maybe they need to have the receipt of the camera that it was taken with but in practice is this really something anyone is ready for. [truncated for AI cost control]