Hi Reader!

As we march sunkissed and bank holiday bound into August, can I ask you to stop a minute and spare a thought for the EU AI Act?

When the high risk obligations were pushed out to December 2027 and August 2028, most people took that as permission to stop paying attention. The thing is, those deadlines that moved were never really ones that were going to impact most built environment firms as most are not high risk. The one that stayed, Article 50, applies to all the ordinary things that we use daily in our working lives. The documents, the images created, the website chatbots. And it doesn't differentiate by company size. It applies to micro businesses right up to global giants.
​
​Article 50 comes into effect from August 2nd as originally planned. Here's what you need to know.
​
From 2 August, there are two separate duties on AI-generated text, and they land on different people.

1. The first is machine-readable marking, Article 50(2). This is the toolmaker's job, not yours. The provider of the generative system, OpenAI, Anthropic, Microsoft, has to mark its output so it's detectable as AI-generated, through watermarking or embedded metadata. From 2 August for new systems, 2 December for ones already on the market. When you draft in Claude or Copilot, that marking is being handled upstream. Nothing for you to worry about doing here.
2. The second is the one that actually touches most of us, Article 50(4). As a deployer, you have to disclose when text is AI-generated, but only when two things are both true: the text is published to inform the public, and it's on a matter of public interest. A blog post on procurement risk, a public consultation note, a statement on cladding safety, those inform the public on public-interest matters and are caught by this rule. A tender for a client, an internal report, a proposal, private correspondence, a fee letter, none of that is 'informing the public', so the text-disclosure duty doesn't bite at all.

Then there is the release valve, and it's a generous one. Even for public-interest text, no disclosure is required if the content had human review and a named person holds editorial responsibility for it. So review it, put a name against it, own it, and the label duty falls away.

One honest caveat. 'Matters of public interest' isn't exhaustively defined in the Act, so the edges are a judgement call. A thought-leadership piece on housing policy is clearly in. A general marketing blog with no policy angle is arguably out. Where a document sits near that line, the safe move is either a light disclosure or a documented review-and-sign-off, and the second is the one that also makes the label unnecessary.

For what it's worth, we have already put a name against everything public we publish, and it turned out to be an afternoon work, not a project.

We're running a webinar in early September on how to set this up in your own workflow, so a named person stands over your public-facing work and the label duty falls away. If you'd like to secure an early place, you can sign up here.​

This week on ChattingGPT: Karen Lewis Enright

Karen Lewis Enright is a change management professional with over 20 years of transformation experience across global organisations including Johnson & Johnson and Wipro. She holds a Masters in organisational psychology, but what makes her stand out is two decades spent helping leaders turn strategy into meaningful change. She specialises in AI strategy and is the first change management professional to join ChattingGPT.

In this latest episode of ChattingGPT, Maryrose and Karen get into the human side of AI adoption:

• How to lead when you're uncertain yourself, using the image of a slipping cup of coffee to explain where people's attention goes under stress
• Why the sceptics belong in the room alongside the enthusiasts
• How to build trust by being honest with people about what AI changes in how they work.

Karen tells the story of a self-parking car that never gets used, a miniature version of every stalled AI rollout, and makes the case for celebrating the curious champions most organisations try to shut down. This week is a good one: Karen is a straight talking AI change management specialist. Take a listen.

🎧 Apple Podcasts

🎧 Spotify​

​Soon we'll be adding Quboz. Here's why. 🍉

What are you doing this Thurs?

If you're about Dublin town, and you're feeling summery, why not join us on the roof of our new offices in Grand Canal Dock for a lunch and learn, or relax and enthuse AI session?

We recently had our Strategy Day there and the chats and hacks and ideas that flowed after the meeting, we feel, are too good to keep to ourselves.

Come along on Thurs 30 July from 12.30pm.
Some of our trainers will be there. Meet the agent builders. More importantly, meet each other and let's have a really good chat about where we're all at with the various tools at this time.

Save your spot: RSVP here​

How would you like your Fridays back?

...And comply with Article 4 requiring employers to ensure staff involved with AI systems have sufficient understanding of AI capabilities, risks, and operations.
​​
Our autumn training starts on 15 September.
​

Our promise to you: invest a little time with us, and we'll give you Fridays back.
The average amount of time we save people who learn with us is 8 hours. That's a whole day.​
We've got specific AI training for built environment professionals and Claude, Copilot and Core for everyone else. You just need to book your place now. Before you go on holidays.
Don't way we didn't warn ya!

​Browse all public courses >>​

Yours, with my name on it,
​Maryrose, AI Institute