Article 5 of Regulation (EU) 2024/1689 is the list of prohibited AI practices: the ones that cannot be placed on the market, put into service or used, with no exception for size or sector.
The Digital Omnibus added two more. And as usually happens when a closed list grows, the first thing to appear was a convenient way of referring to them that isn't the Regulation's.
How many points Article 5(1) has
Eight: from a) to h). Subliminal manipulation, exploiting vulnerabilities, social scoring, predicting crime through profiling, untargeted extraction of facial images, emotion inference in the areas of workplace and education institutions (except for medical or safety reasons), biometric categorisation of sensitive attributes, and real-time remote biometric identification in publicly accessible spaces for law-enforcement purposes.
It's worth a note, because it's easy to miscount: within h) there's its own list with i), ii) and iii) — the purposes that exceptionally permit that identification — and paragraph 2 has its own lettered list. Anyone counting the letters that appear on the article's page will get well past eight. At the top level, in paragraph 1, there are eight.
What the two new ones are called
Regulation (EU) 2026/1744 doesn't add them at the end. Its point 7 says, literally, that “in paragraph 1, the first subparagraph, the following points are inserted”:
(ba)— AI systems that generate or manipulate “realistic images, videos, audio or similar material of an identifiable natural person's intimate parts, or of an identifiable natural person engaged in sexually explicit activities, without that person's freely-given, specific, informed, unambiguous and explicit consent”.(bb)— systems that generate or manipulate material or performance within the meaning of Article 2, points (c) and (e), of Directive 2011/93/EU, except where a “without right” defence applies under national law.
They're inserted, not appended. They go between b) and c), using the usual legislative technique so as not to renumber a list that other rules already refer to. The article also adds two new paragraphs — 1a and 1b — with clarifications for the purposes of those two points.
Why this matters more than it looks
In any organisation working with Article 5, an informal numbering soon shows up: 'category 9', 'the ninth prohibition', 'the new one'. It's convenient because the original list had eight and these arrived afterwards.
There is no letter i) of prohibited practices, and no 'category 9', in the Regulation. It's an internal label — useful among people who share the context, and wrong the moment it leaves that context.
When that numbering crosses over from a conversation into a document, and from there into an email to a client or a reply to an authority, it becomes a citation of something that doesn't exist. And unlike an expired figure, this one isn't fixed by re-reading: the Regulation doesn't contain the reference being checked against, so whoever verifies it doesn't find an error — they find nothing.
It's the same family of problems as an obligation attributed to the wrong party: it starts from a reasonable shortcut and survives because its content is correct. What fails is the reference.
When they apply from
Not from the Omnibus's entry into force. The amended Article 113 expressly carves out these points and the two new paragraphs: they apply from 2 December 2026, while the rest of Chapters I and II keeps applying from 2 February 2025.
Another date of its own within a calendar that already has several — and another reason not to trust a summary that talks about 'the prohibitions' as a block. The full calendar tells them apart milestone by milestone.
What this means for a company that uses AI
For the vast majority, nothing operational: these are practices no legitimate organisation carries out. Article 5 isn't a list of things to do well, it's a list of things you don't do.
Where it does matter is in two specific places.
When assessing generative tools. If your company brings in a system that generates or manipulates image, video or audio, the question of whether it can produce that kind of content — and with what safeguards — stops being theoretical. Not because you're going to use it that way, but because Article 5 reaches “the use”, not just the placing on the market.
And in the precision of what gets documented. If an internal record simply notes that 'category 9' was checked, in two years nobody will know what was actually checked. If it notes that points (ba) and (bb) of Article 5(1), as amended by Regulation (EU) 2026/1744, were reviewed, it will be known exactly — and it can be verified.
That difference costs nothing the day it's written. It costs a great deal the day it has to be reconstructed.
Content in accordance with Article 5 of Regulation (EU) 2024/1689, as amended by Regulation (EU) 2026/1744 (Official Journal of the EU, 24 July 2026).
This article is for informational purposes only and does not constitute legal advice.