Yesterday we wrote that from 2 August, a realistic synthetic presenter in a learning programme carries a disclosure duty under Article 50 of the EU AI Act. It applies whether or not a real person is depicted, and whether or not anyone intended to deceive.
Module 1 of AI for Professional Practice is now live. It has two synthetic presenters. Before either of them speaks, the module says so:
That screen is the whole programme in miniature. The disclosure is a legal requirement. We have made it the opening argument rather than the small print.
What it is
AI for Professional Practice is a CPD programme for regulated professionals. Solicitors, accountants, insurance and financial services professionals, company directors. People who carry explicit professional duties written long before AI existed, and who now use AI most weeks.
Module 1 is called Foundations. About twenty minutes, one sitting, no sign-up to start. It is free. At the end there is a certificate, and it states what you demonstrated rather than that you attended.
It is deliberately not a course about AI. There are plenty of those. It is about four duties you already carry, and what changes when a machine writes the first draft. Accountability, verification, confidentiality, judgement.
The first ninety seconds
Before anything else, the module asks you three questions. One of them is this. When AI drafts something for you, how thoroughly do you check it before your name goes on it?
You answer honestly, because nobody is watching and there is nothing at stake.
Ninety seconds later you are looking at a document that needs your approval. It reads well. It is properly structured, confident in tone, and it cites an authority that supports its central claim.
Most people approve it.
The authority does not exist. Not misquoted, not out of date. Never written by anyone. And if you do not believe that, the module hands you a real search field, pre-filled, and lets you go and look for it yourself. It returns nothing, because there is nothing to return.
Then it shows you your own answer to the question from ninety seconds earlier.
Nothing is said about the gap between the two. It does not need to be. That gap is the module, and every learner does the arithmetic without help.
This is why it opens this way rather than explaining. Being told that AI fabricates sources changes very little. Approving one yourself, with your name on it, at 16:47 on a Tuesday, changes quite a lot.
What you carry out of it
A working routine for checking AI output before it goes out under your name. Four checks: the sources exist, the quotes match, the figures trace, the claims hold. You will have run it once already before it is named.
A clear line on confidentiality. What can and cannot go into a public AI tool, and what the safe alternatives actually are.
A way of making the disclosure call. When to say AI was used, when it does not matter, and when the right professional answer is not to use it at all.
And one sentence worth more than the rest together. You can delegate the drafting. You cannot delegate the authorship.
How it is assessed, and why that matters commercially
The module ends with a single continuous scenario across one working afternoon, and it assesses all four duties inside it. Nothing can be answered by remembering a screen. Every decision is a judgement call with a realistic consequence shown afterwards. Get one wrong and the module tells you what your choice caused, not that you were wrong.
The certificate is issued on the result, at a pass mark of 70. You also get a breakdown by duty, so you can see which of the four you are weakest on, which is more use than a percentage.
One design decision is worth stating openly, because buyers in this sector are beginning to ask about it. The assessment is fixed and written in advance by people. It does not use AI to adapt to the learner or generate questions on the fly. Under Annex III of the AI Act, an AI system that evaluates learning outcomes or determines access to education is classified as high risk, with everything that follows from that. Those obligations were deferred to December 2027 this month. They are still coming. A pre-authored assessment sits on the safe side of that line by design, not by luck.
Keeping it honest
Every module carries a stamp on screen: last verified accurate, next review due. The content is re-checked every quarter.
A course that warns people about unverified output should be able to show its own working. If we ever let that stamp go stale, you are entitled to draw the obvious conclusion.
For professional bodies and training providers
The programme is built to be licensed. Your branding, your platform, your members, with the same quarterly re-verification so it does not quietly go out of date on you. Sector editions can be built on the same chassis for a specific profession.
If that is worth a conversation, the contact page is the place to start. Early conversations shape how the licensing round is structured.
Take it
Module 1 is live now. Module 2, Verification in Depth, follows in September, and there is a waitlist on the programme page.
The most useful thing we can say about it is also the least persuasive. Go and take it, and decide for yourself.
Thinking about a programme of your own rather than a module of ours? The Programme Design Diagnostic is a fixed-fee, independent read on a single programme: a board-ready diagnosis and a costed build scope, in two to three weeks.
For the disclosure duty that applies to synthetic presenters from 2 August, see the AI Act's transparency rules were not delayed. For the enforcement timeline itself, the companion piece is what actually happens on 2 August 2026. For the separate literacy obligation under Article 4, see AI literacy in plain English. See more insights from LearnFrame.