Article 50 of the EU AI Act…
Toward the end of 2024, I rebuilt my site InspirationIsPurple. As the original site had been lacking content, I turned to Hostinger’s new AI content tool to assist in the generation of the introductory posts I wanted to publish. Whilst edited to suit, you could still take an educated guess that the posts were machine generated but they served their purpose. So rather than completely rewrite a dozen or more posts I decided to add my own clumsy AI disclosure that I had assistance. Little did I know at that time, it was about to become a thing.
Fast forward to 2026 and the BMCS rebuild, I actually added a section to my Terms & Conditions of Use: an AI Disclaimer. Not because my huge legal team told me to (Google / AI, of course) but because, with a little research and assistance, I discovered there was an Article 50 of the EU AI Act on its way. So it made sense to pre-empt its enforcement.
The Commission’s Code of Practice has since been finalised, with the final Guidelines adopted alongside it. The requirements of the Act became enforceable on 2 August 2026.
So, I started writing this a couple of weeks after the event. What started out as a “here’s what I did, and it turns out I did the right thing.” article has turned into a “here’s what I did, but I don’t really have to do it.” article. The reason will become clear shortly.
Why I bothered before I had to…
I have been using AI tools — Claude, ChatGPT, Canva — throughout BMCS. Coding assistance, content research, first drafts of articles, image generation. None of that is a secret, and I don’t think it should be. But “not a secret” and “clearly disclosed” are different things, and Article 50 is specifically about the second one.
Article 50 covers several situations, but the one that I believed applied to me as a site owner publishing written content can be summarised as follows: if AI played a meaningful role in producing something a visitor reads, that involvement should be visible, not just technically discoverable if someone digs through a Terms page nobody reads.
I didn’t want to wait for the Guidelines to be finalised to start acting on that. Waiting felt like the wrong instinct when my whole approach is based upon being upfront about how I work.
Where the initial structure actually came from…
Since it reflects the whole point of the article: I asked Claude for guidance on how to structure the disclaimer itself — what a transparent, workable system might look like before I started drafting Terms of Use text and adding an appropriate disclosure.
I didn’t take that at face value. I checked it against the Article 50 draft and the guidance available at the time, adjusted it accordingly based on my own interpretation, and only then added it to the site with the two-tier structure below forming the AI disclosure part of the process.
The two-tier system I implemented…
I didn’t want a single blanket “AI was involved somewhere” notice — it’s not informative, and it doesn’t reflect how differently AI actually gets used across my articles. So I split it into two labels to differentiate:
- AI Generated — the initial draft came from an AI language model. I edited, reviewed, and fact-checked it before publishing, and I hold full editorial responsibility for the final version.
- AI Assisted — I drafted and wrote the article myself, using AI as a tool along the way — exploring ideas, checking facts, refining phrasing. Same editorial responsibility either way.
Both labels carry the same closing line, deliberately: “I hold full editorial responsibility for everything published on this site.” because that is the actual point. AI disclosure wasn’t about diluting accountability, it was about being clear on where the work came from while still owning what gets published.
Where the labels show up on the site depends on the article — some are flagged inline near the top, others in the article footer, depending on the requirement. I kept the exact placement flexible, because the goal was honesty, not decoration.
What the finalised guidance actually confirmed…
Now this is where things got a little messy and resulted in a re-draft of this article. It is worth noting as it highlights two key points:
- The importance of the human review element referred to in the guidelines
- There is a reason why your AI Tool acknowledges it can make mistakes and output should be verified
When I first reviewed the finalised guidance, with Claude’s assistance, I was fairly reassured that the position I’d taken and the process I’d implemented was in line with that guidance and that the following conclusion was one to share:
- The transparency obligation is mandatory — if AI touches published content, disclosure is required, full stop.
As part of my final review (of the initial draft post) I went back and checked some of the Act’s own explanatory text — and discovered I’d been unnecessarily cautious. Recital 134, which sits behind the actual Article 50(4) obligation, reads:
In addition, it is also appropriate to envisage a similar disclosure obligation in relation to AI-generated or manipulated text to the extent it is published with the purpose of informing the public on matters of public interest unless the AI-generated content has undergone a process of human review or editorial control and a natural or legal person holds editorial responsibility for the publication of the content.
The initial interpretation wasn’t entirely correct.
So this is my revised take: most of what I publish here — personal accounts, technical how-tos, opinion pieces — probably doesn’t even count as “informing the public on matters of public interest” in the first place, so the obligation likely doesn’t apply to it at all. And for anything that might arguably fall under that definition, I genuinely try to review and fact-check everything before it goes live, and I hold named editorial responsibility for every word published here. Either way, the obligation appears to be discharged.
Anyhow, I’m not a lawyer, and this isn’t legal advice — but if you’re running something more complex than a standalone content site (an image-generation tool, a chatbot, anything closer to “providing” AI output rather than just using it), the obligations to disclose are very real. So I’d recommend reading the source material directly to understand the requirements rather than relying on secondhand summaries — mine included.
Disclaimer updates…
As I stated earlier I’d rather be transparent about my use of AI. So whilst the obligation is not a requirement, I will continue to include my Terms of Use AI Disclaimer and use my two tier system of AI disclosure, but maybe with an odd tweak to the wording. I’d already added the AI Disclaimer with a caveat that I’ll revisit it as the rules mature and that will continue to be the case. The Code of Practice is still fairly new and I’d rather commit to revisiting it than forget about it because it does not appear to apply to me.
A quick self-check, if you’re working through this too…
If you publish content and use AI anywhere in the process, here’s a suggested rough and ready check:
- Does what you publish actually inform the public on a matter of public interest?
A lot of content — personal accounts, tutorials, opinion pieces, niche hobby writing — probably doesn’t meet that bar, in which case the disclosure obligation likely doesn’t apply to you at all. - If it might qualify, do you genuinely review and fact-check AI-touched content before it publishes?
Not a spellcheck — do you undertake an actual substantive review of what’s been written/created. - Is there a named person or organisation who holds real editorial responsibility, with easy locatable contact details?
If both this and the previous point hold true, you’re likely covered by the exemption even for content that does touch public interest. - Even if you’re exempt, do you want to disclose anyway?
That’s not a legal question — it’s a values one. I’ve decided the answer’s yes for me, but it’s worth deciding deliberately rather than defaulting either way.
PhilA is human intelligence and can make mistakes. Please double-check his information before acting on it.
