On Sunday 2 August the EU AI Act’s transparency rules became enforceable, along with the penalty regime and the authorities that will apply it. The headline that many people are reading is “Brussels delays the AI Act”. That’s only half true. The high-risk obligations are delayed, but the transparency provision isn’t. The EU has also quietly softened the duty to ensure AI literacy. It now simply asks organisations to support AI literacy and competence rather than ensure it.
For Purposeful Relations you might think that’s a blow, as a legal requirement for PR and comms team to do training should have led to lots of new work developing AI strategies and providing the AI training. Undoubtedly it would have done, but for the wrong reason. PR and comms leaders shouldn’t be developing AI strategies and providing AI training to their teams because they fear a draconian fine. They should be doing both because they benefit their team and organisation by helping them to produce better quality work faster.
What actually switched on
I won’t bore you with the precise contents of Regulation (EU) 2026/1744, the Digital Omnibus on AI, which entered into force on 27 July to amend the EU AI Act. The reason for the delay is that by June 2026 just nine of 27 member states were ready to implement it in full.
However, from 2 August, some new duties do apply. Organisations must tell people when they are interacting with an AI system, unless that is obvious from context, and must mark synthetic audio, image, video and text so it is machine-readable. They must inform people exposed to emotion recognition or biometric categorisation, and must disclose deepfakes and AI-generated text published on matters of ‘public interest’ where no human has taken editorial responsibility.
The important qualifier is ‘public interest’. This does not mean anything the public might find interesting, or every blog post, LinkedIn update or marketing article that touches on a topical issue. It means content published to inform the public on matters such as politics, public health, security, the environment, consumer safety, science, culture or other issues of public debate.
Even then, the AI Act’s disclosure duty for AI-generated text is narrow because it does not apply where there has been substantive human review and clear editorial responsibility. In practical terms, a communications team using AI to help draft, structure or challenge a piece should not normally need a transparency label if a named person has checked the facts, exercised judgement, approved the argument and taken responsibility for publication. However, there is a new need to have a procedure in place to ensure the checks take place and they are logged in some way.
The reason so many people are panicking about this is the same as the over-reaction to the original introduction of GDPR. It is the potential draconian fines which means many organisations are taking an overly conservative approach. Breaches of the new transparency disclosure sit in the middle penalty tier, which is still incredibly expensive. It is up to €15 million or 3% of worldwide annual turnover, whichever is higher.
The EU AI Act is extraterritorial. If your output reaches users in the EU, you are in scope. This means UK PR agencies, public affairs consultancies and in-house communications teams do not get a free pass if they work with European clients or their work is directed at Europe.
The literacy duty got weaker. That is the trap.
Article 4 of the EU AI Act has already been in force since 2 February 2025. It required deployers of AI to ensure, to their best extent, a sufficient level of AI literacy among staff and anyone operating AI on their behalf. The omnibus changed the verb to water it down. The duty is now to take measures to support the development of AI literacy. It requires organisations to show they are trying, rather than the result that they’ve succeeded.
There are undoubtedly some relieved procurement teams out there. But I think they are reading it wrong.
Article 4 never carried a fine of its own. Its force was always evidential. This means that when something goes wrong – such as a discrimination claim, a tribunal, a regulatory finding, a copyright breach, a client dispute over a fabricated statistic – the question is what measures did you take. The answer is either documented evidence of policy, procedures and training, or an awkward silence. Softening the standard does not remove the question.
The trap is that developing an AI comms strategy and doing training are delayed again. The real cost of getting AI wrong was always operational and reputational, not the risk of a fine.
The case for AI literacy now has to stand on its own merits. This is good news as it means PR and comms leaders need to invest in training and strategy to reap the benefits and opportunities, not just because they legally have to.
The PR and comms AI readiness gap is real
The evidence from corporate affairs is consistent and alarming. BCG’s 2026 survey of more than 200 corporate affairs and communications leaders at large US and UK companies found that 68% described their function as an AI laggard, while 88% said they were not fully prepared to lead an AI transformation in their function.
That is not a marginal capability gap. Corporate affairs should be helping to ensure that the organisation overall isn’t negatively impacted by AI failures, which it will find impossible if it hasn’t even kept up itself.
The same BCG research found that only 31% had made meaningful progress in scaling generative AI beyond pilots, while most remained in experimentation mode and were seeing little or no value. The top barrier was not fear of AI, but organisational design: 35% cited a lack of operating model design capability, ahead of lack of skills and lack of budget.
The 2026 Corporate Affairs Report from Deloitte found that just a quarter of corporate affairs leaders have a formal AI strategy. This is despite the fact that most answered AI when asked to complete the sentence “2026 is going to be the year of….”.
That matters because corporate affairs and communications rank among the top two enterprise functions for generative AI transformation upside, with BCG estimating productivity gains of roughly one-third or more across critical tasks and processes. The gap is leadership, operating model and adoption discipline.
The gap is not an AI tools problem as AI licences are cheap and many organisations have already invested in them, but aren’t seeing the benefits they’d hoped for. We call it the AI adoption illusion: the belief that buying the technology delivers the transformation. It does not. Impact comes from changing how people relate to the tool — strategy, governance, culture, structured training, sustained reinforcement.
Transparency is a symptom, not a strategy
The new transparency requirements are being sold as this month’s compliance scramble. It should be the least of your worries, provided the foundations are in place.
Think about what the duties actually demand. To label a deepfake, someone has to recognise one. To disclose AI-generated text on a matter of public interest, someone has to know what human editorial responsibility means in practice. To flag a chatbot, someone has to have mapped where chatbots sit in the customer journey. It requires everyone to understand what ‘public interest’ means and when it applies. Every disclosure obligation resolves to a person making a judgement about a system they understand.
Which is why we treat transparency as the output of good governance, not a workstream of its own. Get the inventory, policy, accountability and training right, and labelling falls out of the process almost automatically. Get them wrong and you are applying stickers to content nobody can vouch for. That is compliance theatre, and it fails at the moment it matters.
There is a professional dimension too. Transparency, disclosure and authenticity are the first of the ICCO Warsaw Principles, ratified by our industry three years before Brussels legislated. In June the CIPR has published a best practice guide on translating Global Alliance’s Venice Pledge into public relations practice. In February transparency is one of the themes in the PRCA Green Paper I co-authored on the responsible use of AI in public relations and communications.
If you needed the AI Act to tell you to disclose synthetic content, regulation is not your biggest problem.
What to do in the next 90 days
If you are a PR or communications leader, the five things you need to do in the next 90 days are:
AI Governance: Establish operational AI governance. Do not rely on an IT or legal policy—neither team understands external reputational risk or workflow integration.
AI Strategy: Build a timeline covering team adoption, enterprise-wide reputational guardrails, social licence, and AI as an active stakeholder.
AI Planning: Audit workflows, tasks, and capability gaps to create a pragmatic adoption roadmap.
AI Training: Implement a structured learning program tailored to the specific risks and tools identified in your audit.
AI as a stakeholder: Audit how generative engines describe your organization and upskill your team on Generative Engine Optimisation (GEO).
Get in touch with us today to find out how our proprietary CommsTransform™ IDEA framework and AI Stakeholder Framework™ can help.
Transparency disclosure – Written by me. Checked by Microsoft 365 Copilot. Tweaked by me based on Copilot’s criticisms, but its suggestions often rejected. It is better at identifying what is wrong, but I’m better at fixing it!
