EU AI Act Officially Enforced Today: Chatbots Must Disclose AI Identity, Deepfake Content Mandatory Labeling

Under the new rules, chatbots and other interactive AI systems must clearly inform users that they are interacting with AI rather than humans. This requirement may seem simple, but in practice involves numerous technical details. For example, when users consult through a customer service phone system, the system must clearly inform users at the start of the interaction that they are speaking with an AI assistant, rather than disclosing this during or after the interaction. For deepfake content — including images, videos, or audio edited or generated using AI — clear labeling is mandatory. More importantly, AI-generated or modified content must also carry machine-readable marks so that automated systems can more easily detect and identify this content. The purpose of these measures is to reduce deception and manipulation, help people make informed choices, while also providing businesses with clearer compliance obligations and practical methods to demonstrate compliance.

Notably, the 'high-risk AI systems' provisions originally scheduled to take effect today have been postponed. Under revisions in the Digital Omnibus package, Annex III high-risk AI system compliance requirements — covering recruitment tools, credit scoring, education, law enforcement, border control, and critical infrastructure — originally set for full enforcement on August 2, 2026, are now delayed to December 2, 2027, an extension of 17 months. Legislators recognize that compliance requirements can only be effective when necessary infrastructure is in place. This adjustment is not simply granting more time, but aims to align rule implementation with the actual availability of systems and structures needed to support the rules, ensuring more coherent and legally certain implementation across the EU. However, the AI Office's supervisory powers over general-purpose AI models (GPAI) have been strengthened — for vertically integrated providers (which describes most frontier labs), supervisory authority is centralized in Brussels rather than dispersed across 27 national regulators.

For generative AI systems already on the market, legislators introduced a four-month 'grace period.' The watermarking obligation under Article 50(2) of the AI Act was originally set to apply from August 2, but for generative AI systems already on the market before August 2, 2026, the compliance deadline is extended to December 2, 2026. This arrangement gives existing AI service providers a buffer period to implement watermarking technology. Simultaneously, the AI Act introduces a new prohibited practice: banning AI systems designed to generate non-consensual intimate imagery (so-called 'nudifier' applications) and child sexual abuse material. This ban takes effect on December 2, 2026, representing an important addition to the existing prohibited practices list. The European Parliament pushed for this provision during trilogue negotiations, responding to concerns about online harms caused by 'nudifier' tools and intimate deepfakes.

The global AI industry's reaction to these new regulations is mixed. Large tech companies like Google, Microsoft, and Meta mostly expressed support, believing clear regulatory frameworks help build public trust and promote healthy industry development. However, some startups and open-source AI communities expressed concerns that compliance costs may impose excessive burdens on smaller businesses. The European Commission responded that it is developing simplified compliance guidelines for SMEs and startups. In the United States, federal AI legislation is also accelerating, but the two parties still have significant disagreements on AI regulation direction. Meanwhile, China is also actively improving its AI regulatory framework. Global AI regulation is forming three major systems — the 'EU model,' 'US model,' and 'China model' — posing significant challenges to multinational AI companies' compliance strategies. For AI companies operating in the EU, today's regulation taking effect means they must immediately adjust products and services, or face fines of up to 6% of global annual revenue.