Real or Really Close? The EU’s New AI Labeling Regime and the Risk of Realistic Deepfakes
What does it mean when a politician’s gaffe is exposed as a fabrication? How is a consumer to trust a commending product review when its penmanship originates from the oft imperceptible hand of an algorithm? In a bid to answer these questions, the European Union has enacted a set of transparency rules under its landmark AI Act, compelling companies to label artificially generated or manipulated content designed to look authentic. Starting from this month, the age of the invisible deepfake is, in principle, over.
The initiative comes as a direct response to the flood of deceptive content that has plagued the information ecosystem in recent years: from fabricated audio of Slovakian opposition leaders conspiring to rig elections to videos of Emmanuel Macron in a 1980s nightclub. The new rules, which came into force on the 2 August 2026, mandate that providers of interactive AI systems such as chatbots ensure users are aware that they are not communicating with a human. More significantly, deployers, (the publishers and companies using AI), must clearly label deepfakes and AI-generated text on matters of public interest if it has not undergone substantive human editorial oversight; this includes politics, public health or the environment. This is an attempt to preserve the integrity of democratic discourse through preventing the manipulation of facts.
I can report that the EU legal framework is more intricate than first appears. The laws create a layered set of obligations that apply not only to high-risk AI systems, but more broadly to any system that interacts with people or generates content. For example, Article 50 requires providers of generative AI systems to embed digital watermarks in their outputs to ensure they are technically detectable as AI-generated. Providers of systems already on the market are given a grace period until the 2 December 2026 to comply with this technical and logistical challenge. The onus falls on the deployer under Article 50, whereby they must ensure this mark translates into a clear and perceivable label for the user- a label which cannot be hidden in a website’s terms and conditions.
While the intent to empower consumers is commendable, the rules have sparked controversies within the tech industry. The most prominent concern, as voiced by the Computer and Communications Industry Association, (CCIA), is that the European Commission’s implementing guidelines have expanded the definition of a deepfake far beyond what was originally envisioned. The concern is that the requirement to label is so broad that it will capture almost everything, turning a benign AI-generated landscape in an advert into the equivalent of a malignant, manipulated political speech. Boniface de Champris, the CCIA’s Ai policy lead, warns that this risks creating a situation similar to that of cookie banners, that once labels are everywhere then users will simply stop noticing them. This undermines the very goal of transparency. Additionally, this fear of legal fatigue is accompanied by the complexity of the definitions and what content requires a label. For businesses, it is my opinion that the question becomes a legal minefield. Where is the line between standard AI-assisted editing, such as color correction or background removal, which is exempt from labeling and a substantive manipulation that requires prominent disclosure? This legal uncertainty creates a risk of ‘over-compliance’, as companies fearing the up to €15 million fines, may label content indiscriminately. The rules offer exemption for creative, satirical and fictional works, a distinction that is crucial yet inherently subjective for businesses to interpret.
So, does, in fact, the EU’s new AI transparency regime signal the rise of a new era of digital accountability, where citizens and consumers alike are armed with the information they need to navigate an increasingly synthetic online world? Will it prove to be well-intentioned, or will it be an ultimately counterproductive measure, drowning the public in a sea of labels and stifling innovation with a one-size-fits-all requirement? Those two things can be simultaneously true at once. The success of this landmark legislation will be determined not simply by its implementation in the coming months, but by whether its labels serve to illuminate the truth or simply become more digital noise.
© Lawrence Power 2026
Whitestone Law
References:
Gates, T. (2026, August 04). Always Read the Label: EU AI Transparency Rules and Questions That Remain Unanswered. Accessible at:
Hamit, D. (2026, August 03). EU rules requiring labels on AI-generated content take effect. Anadolu Ajansi. Accessible at:
Machin, E. (2026, August 03). You Talkin’ To Me? Operationalising The EU AI Act’s Transparency Obligations. Ropes Gray. Accessible at:
Rankin, J. (2026, July 31). AI labels to be compulsory on authentic-looking content under EU rules. The Guardian. Accessible at:
WAM (2026, August 03). EU enforced mandatory labelling of AI-generated content. Gulf News. Accessible at:
(2026, 14 May). The EU AI Act’s Transparency Rules: A Practical Guide to Article 50. Accessible at: https://artificialintelligenceact.eu/transparency-rules-article-50/
(2026, 31 July). Code of Practice on Transparency of AI-generated Content. Europa. Accessible at:
https://digital-strategy.ec.europa.eu/en/policies/code-practice-ai-generated-content