The EU AI Act and what it means for internal communications

New AI rules landed on 2 August. Few have written about what they mean for internal communication. So, here’s what we know...

Search for what the EU AI Act means for internal communication, and there’s not a lot out there. Search for what it means for marketing, and you’ll find commentary to fill your bedtime reading for at least a fortnight. The gap matters, because the parts of the Act that reach furthest into an IC team’s work carry the heaviest fines.

Planning season is the right moment to look into it. Most of what follows won’t change a word of what you publish. But some of it will change what you commission, and one part in particular may mean switching off a feature in software your organisation already owns.

We’re communicators, not lawyers. Anything with fines attached deserves a proper legal opinion. Every claim we make below is sourced, and the sources are listed at the end. 

What changed on 2 August

The Act now says people should be told when they’re interacting with AI.

That covers four things: 

1. Anything that holds a conversation with you

2. Content generated by AI

3. Systems that read emotions

4. Realistic fake video or audio

The European Commission published its guidance on how it all works on 20 July 2026. The relevant section is Article 50. [1][2]

Responsibility is divided into two. Companies that build AI tools have one set of duties. Organisations that use those tools have another. In the Act’s language, you're a “deployer”, and the word is worth recognising because it appears everywhere. The invisible watermarking everyone has been discussing belongs to the toolmakers, not to you. Tools already on sale before 2 August have until 2 December to add it. [3][4]

Which leaves the question most of you will be asking: does any of this impact me? It does, if you use the output inside the EU. That includes European employees, no matter where your head office is.

Nearly everything you publish is a-okay

The labelling rule for written content only applies to text published to inform the public about matters of public interest. An all-staff email isn’t that. Neither is an intranet story, a manager toolkit or a leadership briefing pack. So, everyday IC work is largely untouched. [6]

But, the story doesn’t end there. The Commission defines “public interest” more broadly than news, including financial, political, scientific and cultural matters that could reasonably end up being publicly debated. [4] A pensions message or a restructure announcement sits nearer that line than a story about a new water cooler supplier. 

There’s also an exemption for content a human has properly reviewed and taken responsibility for. That is a genuinely useful exemption, but the guidance sets a bar for what counts as review. Ticking a sign-off box doesn’t cover it. Someone – say a named author – has to own the words.

“It’s internal” won’t hold on its own

There's an exemption for internal use, and it’s pretty narrow. The law firm Bird & Bird has gone through the guidance line by line and pointed out that this only works when two things are true at once: the content is strictly technical, and it only ever reaches a defined group of people inside the organisation. The moment it goes to anyone outside – like a supplier, a contractor, a customer – the exemption falls away. So, saying you only use it internally isn’t enough on its own. [8]

We all know internal comms leaks. So best to continue planning on that basis.

Video and voice are risky business

If next year’s plan includes an AI version of a leader on screen, or a synthetic voice reading a training module, you need to pay close attention.

The rules cover any content realistic enough that someone might take it as genuine. Whether you meant to mislead anyone is irrelevant. Good intentions are no defence, m’lord! This applies to lifelike people who don’t exist at all, not just lookalikes of real people. The guidance asks you to think about everyone who might see it, including children and people less confident with technology. [7]

Where you put the ‘this is AI’ label matters as much as whether you have one. It must sit on the content itself, rather than in the file’s hidden data or a caption someone can crop off. [9] That makes it a storyboard decision, not something to sort out at upload.

If you have a chatbot

Anything that holds a conversation with people must make it clear that it’s AI, unless it’s obvious. [6] 

The Commission borrows a test from consumer law and asks you to consider who’s actually using the thing, how familiar they are with AI, and whether anyone using it might be vulnerable. [7] A workforce spread across depots, wards and shop floors may not be as confident with technology as someone who works on a screen all day every day. People also need to be told at the moment they come into contact with the chatbot, not somewhere in a policy nobody opens. [10]

The rule that could actually get you fined

This one’s a little bit older, and a lot stricter. Using AI to work out how employees are feeling is banned outright across the EU, with narrow exceptions for medical and safety reasons. It has been illegal since 2 February 2025. The reference, if you need it, is Article 5(1)(f). [11][15]

The definition of “workplace” is broad. It covers any setting where work happens and extends to recruitment. Fines here reach €35 million or 7% of global turnover, the highest band in the Act. [11]

The ban covers emotions worked out from physical signals such as faces and voices. Analysis that doesn’t use those signals falls outside it. [12] So running sentiment analysis across written survey comments is a different question from scoring someone’s tone on a video call. Regulators were already on this well before the Act arrived. In 2022, the Hungarian data protection authority found a bank’s use of AI to read emotion in recorded calls breached data protection law and ordered it to stop. [12]

Almost nobody sets out to buy emotion-reading software. It arrives bundled up in a cosy tech blanket. Then a meeting assistant rolls out with engagement scoring turned on, IT approves it, and nobody in the chain thinks of it as an AI decision. Top tip: go and have a look at what your existing tools might actually be doing.

How to stick to the law 

Some of the below is legally required. Some of it’s a judgement call.

Required wherever your output reaches the EU

• Label synthetic video and voice. Anything realistic enough to be taken as genuine needs a visible label on the content itself.  

• Say when a chatbot is a chatbot. Anything that holds a conversation with employees has to make that clear from the start. 

• Check that nothing in your stack is reading employee emotions. The ban on inferring feelings from faces and voices is absolute, and the fines sit in the Act's highest band. This one lives in procurement, so the conversation is with IT rather than with your team.

• Take responsibility for anything touching public interest. If AI helped draft something that could end up in public debate, like pensions, a restructure, or an environmental incident, then a named person needs to have reviewed it and be prepared to stand behind it. That review is what removes the labelling requirement.

  
  

Your call

• Naming an owner for everyday internal content isn’t a requirement. Most of what an IC team publishes never comes near the public-interest test. However, plenty of organisations will want to instil the habit anyway, given how often internal content travels beyond the walls of the organisation.

• Telling people when AI helped write something. An intranet story or a manager toolkit can be AI-assisted and unlabelled, entirely lawfully. If you decide to label anyway, the Commission has already published a set of EU icons for that purpose, free to use and needing no attribution, so there is no point in designing your own. That choice says more about the kind of organisation you are than any of the requirements above.

Planning season is underway, and this is exactly the sort of thing that gets missed. If you'd like a second pair of eyes on next year's plan, we’re here to help. Drop us a note at info@abcomm.co.uk.

Sources

All links verified August 2026.

1. European Commission, Guidelines on transparency obligations for providers and deployers of certain AI systems (adopted 20 July 2026) — https://digital-strategy.ec.europa.eu/en/policies/guidelines-transparency-ai-generated-content 

2. European Commission, Quick Facts: Transparency rules for AI systems — https://digital-strategy.ec.europa.eu/en/factpages/quick-facts-transparency-rules-ai-systems 

3. Cooley LLP, EU AI Act: Transparency Obligations Take Effect 2 August 2026 — https://www.cooley.com/news/insight/2026/2026-08-03-eu-ai-act-transparency-obligations-take-effect-2-august-2026 

4. European Commission, FAQ: Transparency obligations under Article 50 of the AI Act — https://digital-strategy.ec.europa.eu/en/faqs/transparency-obligations-under-article-50-ai-act 

5. Stibbe, The AI Act’s Transparency Obligations: Rules, Scope and Timeline — https://www.stibbe.com/publications-and-insights/the-ai-acts-transparency-obligations-rules-scope-and-timeline 

6. Regulation (EU) 2024/1689, Article 50, via the European Commission AI Act Service Desk — https://ai-act-service-desk.ec.europa.eu/en/ai-act/article-50 

7. Covington & Burling, 10 Takeaways: European Commission Draft Guidelines on AI Transparency under the EU AI Act — https://www.globalpolicywatch.com/2026/05/10-takeaways-european-commission-draft-guidelines-on-ai-transparency-under-the-eu-ai-act/ 

8. Bird & Bird, Taking the EU AI Act to Practice: Reading the Commission’s Draft Article 50 Guidelines — https://www.twobirds.com/en/insights/2026/taking-the-eu-ai-act-to-practice-reading-the-commissions-draft-article-50-guidelines 

9. activeMind.legal, AI transparency requirements in marketing — https://www.activemind.legal/guides/ai-transparency-marketing/ 

10. Addleshaw Goddard, AI transparency under the AI Act: what businesses need to know before 2 August 2026 — https://www.addleshawgoddard.com/en/insights/insights-briefings/2026/technology/ai-transparency-ai-act-what-businesses-need-know-before-2-august-2026/ 

11. Oliver Patel, Emotion recognition and the EU AI Act, citing European Commission guidelines on prohibited practices — https://oliverpatel.substack.com/p/emotion-recognition-and-the-eu-ai 

12. Future of Privacy Forum, Red Lines under the EU AI Act: the prohibition of emotion recognition in the workplace and education institutions — https://fpf.org/blog/red-lines-under-eu-ai-act-unpacking-the-prohibition-of-emotion-recognition-in-the-workplace-and-education-institutions/ 

13. European Commission, Code of Practice on Transparency of AI-generated Content — https://digital-strategy.ec.europa.eu/en/policies/code-practice-ai-generated-content 

14. Bird & Bird, Taking the EU AI Act to Practice: Understanding the Draft Transparency Code of Practice — https://www.twobirds.com/en/insights/2026/taking-the-eu-ai-act-to-practice-understanding-the-draft-transparency-code-of-practice 

15. Wolters Kluwer, The Prohibition of AI Emotion Recognition Technologies in the Workplace under the AI Act — https://legalblogs.wolterskluwer.com/global-workplace-law-and-policy/the-prohibition-of-ai-emotion-recognition-technologies-in-the-workplace-under-the-ai-act/