AI governance has moved into the publishing workflow
The EU AI Act’s transparency requirements have applied since 2 August 2026.
For marketing, communications and public-sector teams, this is not simply a legal or technology update. It changes how we approve content, use AI-generated material, manage chatbots and record editorial decisions.
My advice is not to respond by labelling everything created with the help of AI. That would create noise and confusion.
The real task is to understand which content crosses the legal threshold, when human editorial review provides an exemption, who is responsible for the final publication and what evidence your organisation should retain.
The European Commission published its final Article 50 transparency guidelines on 20 July 2026. The guidelines are not legislation, but they explain how the Commission expects the binding transparency requirements to be interpreted and enforced.
What Article 50 means for communications teams
There are four areas communications leaders need to understand.
First, providers of AI systems that interact directly with people must make it clear that the person is interacting with AI, unless this is already obvious. This includes conversational AI systems, agents and avatars capable of a genuine two-way exchange.
The notification should appear at the beginning of the first interaction. It must be clear, distinguishable and accessible.
If your organisation commissions, brands or puts an AI chatbot into service under its own name, do not automatically assume you are simply using somebody else’s technology. Your organisation’s status as a provider or deployer should be assessed during procurement.
Second, professional users of AI must disclose image, audio or video content that meets the legal definition of a deepfake.
The Commission describes a deepfake as AI-generated or manipulated content that closely resembles an existing or plausibly existing person, object, place, organisation or event and could falsely appear authentic or truthful.
The disclosure must be visible or audible when someone first encounters the content. Hidden metadata or a machine-readable watermark is not enough.
Third, AI-generated or manipulated text published to inform the public about matters of public interest may need to be labelled.
The Commission’s examples of public-interest matters include politics, public administration, public services, public health, public safety, fundamental rights, environmental protection and economic, financial, scientific or cultural developments that could reasonably form part of public debate.
This is a broad category. It will capture a significant amount of content published by government departments, local authorities, public bodies, political institutions and the agencies working for them.
The fourth area concerns provider-side machine-readable marking. Providers of generative AI systems must ensure that qualifying AI-generated or manipulated content can be detected electronically.
This is separate from the responsibility of the organisation publishing the content. A technical watermark does not remove your obligation to decide whether a visible disclosure is required.
The Commission’s detailed Article 50 questions and answers confirm that the requirements are not retrospective. Content generated and published before 2 August 2026 does not have to be relabelled.
Human review must now mean something
The human-review exemption is especially important for communications teams.
AI-generated or manipulated public-interest text does not have to be labelled when it has undergone substantive human review or editorial control and a person or organisation accepts editorial responsibility.
However, running an AI draft through a spelling or grammar checker is not substantive human review.
The Commission describes human review as a deliberate examination of the substance by someone with relevant knowledge and professional judgement. Editorial control means having the authority to approve, change or reject the content, including checking factual claims and the reliability of sources.
Someone must also accept ultimate legal responsibility for publication.
This provides a workable route for responsible AI-assisted communications. You can use AI to help draft an article, public-information notice or policy explainer without necessarily labelling the finished text, provided a competent person genuinely reviews the substance, verifies the sources and accepts editorial responsibility.
My advice is to document that process.
Record who reviewed the content, which sources were checked, whether significant changes were made, whether a disclosure was considered and who approved publication.
A named editor and an approval record provide much stronger evidence than a policy stating that a “human was in the loop”.
The Commission also confirms that fines for breaches of these transparency requirements can reach €15 million or 3 per cent of worldwide annual turnover, subject to the applicable proportionality provisions.
The Digital Omnibus has changed the timetable
The Digital Omnibus is no longer a proposal. Regulation (EU) 2026/1744 was published in the Official Journal on 24 July and entered into force on 27 July 2026.
The revised dates are:
- 2 December 2027 for high-risk systems listed in Annex III.
- 2 August 2028 for high-risk AI embedded in regulated products.
- 2 December 2026 for providers of generative AI systems placed on the market before 2 August 2026 to implement the machine-readable marking requirement.
That final grace period applies to the provider’s technical marking obligation. It does not postpone the communications team’s duty to disclose deepfakes or qualifying public-interest text.
The Omnibus has also changed the AI literacy requirement.
Organisations must take measures to support the development of AI literacy among staff and other people using AI on their behalf. They do not have to guarantee that every individual reaches a particular level.
The Commission’s updated AI literacy guidance confirms that there is no prescribed certificate, mandatory training format, AI officer or governance board.
This does not mean the AI literacy obligation has disappeared. Organisations still need proportionate, role-specific measures and evidence of the training, guidance or support they provide.
Ireland now has its national enforcement legislation
President Catherine Connolly signed the Regulation of Artificial Intelligence Bill on 21 July 2026, meaning it became Irish law. The President’s notice confirms the enactment.
The legislation establishes Oifig IS na hÉireann, the AI Office of Ireland, as the central coordinating authority for implementation of the AI Act.
Ireland will operate a distributed enforcement model. Existing sectoral regulators will supervise AI within their areas of responsibility, while the new office will coordinate implementation and act as a central contact point.
The Government’s description of the legislation includes investigation, compliance, prohibition, seizure and sanctions powers.
Different domestic provisions may require commencement through ministerial orders. Organisations should therefore obtain legal advice before relying on a particular Irish enforcement power, regulator or commencement date.
The directly applicable EU requirements are not suspended while Ireland completes its domestic implementation arrangements.
Google and Meta have signed the voluntary code
Google announced on 24 July that it was signing the EU Code of Practice on Transparency of AI-Generated Content. Meta followed on 28 July.
Google’s announcement refers to its work on SynthID and C2PA-based provenance. Meta’s announcement commits it to working with the AI Office on practical and interoperable transparency measures.
The code is voluntary. The underlying Article 50 requirements are binding.
These announcements do not remove the responsibility of advertisers, publishers or communications teams to assess their own content. Platform labels should support your compliance workflow, not replace it.
Here is what you need to act on now
1. Add an Article 50 check to your publishing process
Before publishing, determine whether the asset is a deepfake, whether the text informs the public about a matter of public interest and whether a visible disclosure is required.
2. Name the person accepting editorial responsibility
Identify who has the authority to approve, change or reject AI-assisted content. Do not rely on a vague statement that a human reviewed it.
3. Create an AI-assisted content record
Record the reviewer, sources checked, approval date, changes made and the decision to apply or omit a label.
4. Update staff training
Remove any wording claiming that every employee must reach a guaranteed or formally certified level of AI literacy. Training should be practical, role-specific and recorded.
5. Review your chatbots, avatars and AI agents
Confirm that users are informed at the beginning of the interaction. Assess whether your organisation could be considered the provider because the system operates under your name or brand.
6. Update policies and supplier agreements
Separate provider-side machine-readable marking from deployer-side public disclosure. Clarify the responsibilities of agencies, contractors and freelancers working under your organisation’s authority.
7. Correct your presentations and published guidance
Replace outdated AI Act implementation dates. Update any website, policy or training content that still presents AI literacy as requiring a guaranteed level for every individual.
8. Strengthen your editorial workflow
If you publish AI-assisted articles, reports or public-information material, require substantive human review and retain the primary sources used. A quick proofread is not sufficient.
9. Keep platform rules separate from legal duties
Record platform labelling and disclosure changes in your publishing checklist, but do not treat a platform’s automated label as proof that your organisation has complied with Article 50.
10. Get legal review for borderline decisions
Seek legal advice before relying on the human-review exemption, deciding that realistic synthetic content is not a deepfake or omitting a disclosure from public-interest material.
Complete the AI Rulebook readiness check to identify where your content, AI systems, staff training and governance processes may need attention.
This article provides general information and is not legal advice. Obtain legal review for your organisation’s role, disclosure decisions, regulatory obligations and reliance on statutory exemptions.
About the Author
AI and Communications Specialist | Author | Podcaster | International Speaker
Joanne Sweeney is a recognised leader in digital marketing and digital transformation, specialising in the practical application of AI to enhance communication strategies. The founder of Digital Training Institute, Joanne empowers leaders and communications professionals to leverage AI responsibly and effectively, ensuring transparency, trust, and tangible results.
She completed the Harvard Kennedy School Executive Programme in AI and Public Policy this summer before launching her fourth book AI Playbook, in September. Joanne has also published two editions of Public Sector Marketing Pro, which were preceded by her first title, Social Media Under Investigation Law Enforcement and the Social Web.
Her books have become go-to resources for governments, public sector and policing agencies, as well as marketing and communications professionals, all navigating the challenges of the changing Internet.
Joanne is the host of the AI SIX Podcast, a daily six-minute podcast that provides actionable insights, expert interviews, and practical advice to help professionals navigate the new world of work.
