The European Union’s AI news this month is being told as a simple story: as of August 2, AI-generated content must be disclosed. That much is true, and it is also the smaller half of what happened. Six days before that deadline, a separate law took effect that delayed the AI Act’s toughest requirements, the ones governing high-risk uses such as hiring, credit scoring and biometric identification, by more than a year. The rules that survived untouched are the ones that reach ordinary AI users: chatbots, deepfakes and undisclosed AI-generated news content. The rules that got pushed back are the ones the industry fought hardest to delay.
Regulation (EU) 2026/1744, the Digital Omnibus on AI, entered into force on July 27 after the European Parliament adopted it on June 16 and the Council signed off on June 29. It amends the original 2024 AI Act to push back Chapter III’s high-risk system rules: obligations tied to Annex III use cases, including biometric identification, employment, credit and education, now apply from December 2, 2027, and obligations for AI systems embedded in regulated products such as machinery and medical devices move to August 2, 2028. Both dates land more than a year later than originally scheduled. The Omnibus’s own recitals give the Commission’s rationale: preparation of the technical standards providers need to demonstrate compliance had fallen behind schedule, leaving “a compliance burden that is heavier than expected.”
The disclosure duties in Article 50 were not part of that delay. Providers must still design AI systems so people know when they are talking to a chatbot, unless that is already obvious. Generative AI outputs need machine-readable markers, with an exception for tools performing only “assistive” editing. And deployers who use AI to generate or manipulate a deepfake depicting a real, identifiable person, or to publish AI-generated text on public-interest topics without an editor’s sign-off, must disclose that AI was involved, with a lighter version of the duty for satirical, fictional or artistic content. The law frames these duties in terms of “deployers” generally, not platforms specifically, though the primary text available does not spell out where that category’s boundaries sit for an individual creator using a third-party AI tool.
The Omnibus did add a grace period on the marking requirement, but only for systems already on the market before August 2, giving those providers until December 2, 2026, to comply. The deepfake and chatbot disclosure duties carry no such grace period.
How Advertisers and Rights Groups Read the Guidelines
The European Commission published guidelines on July 20 to help providers and deployers apply Article 50, alongside a voluntary Code of Practice the Commission and the EU’s AI Board have endorsed as an adequate compliance route. “With today’s guidelines, the Commission supports the smooth and effective application of the AI Act to make AI systems interacting with people, such as chatbots and AI agents and AI content, more transparent and trustworthy,” said Henna Virkkunen, the Commission’s EVP for Tech Sovereignty, Security and Democracy.
The European Advertising Standards Alliance, the self-regulatory body for European ad standards, welcomed the guidelines’ effort to define “deep fake” but said the Commission’s own guidance strayed from the law it interprets, “regretting a tendency in the Guidelines to extend the concept of ‘deep fake’ beyond the wording of the AI Act.” EASA separately welcomed the guidance’s recognition that some AI-generated ad content can qualify for a lighter disclosure regime as creative or satirical work.
A coalition including European Digital Rights (EDRi), Access Now, the European Center for Not-for-Profit Law (ECNL) and Amnesty International objected to the Omnibus itself rather than to the disclosure rules. The groups said the law “delays key protections, weakens transparency and creates a dangerous precedent,” pointing specifically to a change that lets providers upload less information to the EU’s public AI database when they classify their own systems as not high-risk, a change the groups argue limits regulators’ and researchers’ ability to check those self-classifications.
A Separate, Older Rule on Training Data
A less-discussed AI Act provision has already been in force for a year. Article 53 has required providers of general-purpose AI models, since August 2025, to publish a summary of the content used to train their models, maintain technical documentation, and put in place a policy for complying with EU copyright law, including honoring rights reservations set out under the EU’s 2019 copyright directive.
A coalition of European creative-industry associations, including the International Confederation of Societies of Authors and Composers (CISAC), the European Grouping of Societies of Authors and Composers (GESAC), the International Federation of the Phonographic Industry (IFPI) and News Media Europe, welcomed the provision back in March 2024, when the AI Act cleared Parliament, calling it a first step toward letting rightsholders enforce their rights against AI developers and pressing for direct involvement in setting the technical template for those training-data summaries. That statement predates this month’s disclosure deadline by more than two years, and no rightsholder claim or enforcement action tied to Article 53 has been reported since it took effect.
The guidelines meant to help providers meet the disclosure duties that did take effect on schedule arrived just 13 days before the deadline.
Dragomir is a Serbian freelance blog writer and translator. He is passionate about covering insightful stories and exploring topics such as influencer marketing, the creator economy, technology, business, and cyber fraud.
The European Union’s AI news this month is being told as a simple story: as of August 2, AI-generated content must be disclosed. That much is true, and it is also the smaller half of what happened. Six days before that deadline, a separate law took effect that delayed the AI Act’s toughest requirements, the ones governing high-risk uses such as hiring, credit scoring and biometric identification, by more than a year. The rules that survived untouched are the ones that reach ordinary AI users: chatbots, deepfakes and undisclosed AI-generated news content. The rules that got pushed back are the ones the industry fought hardest to delay.
Regulation (EU) 2026/1744, the Digital Omnibus on AI, entered into force on July 27 after the European Parliament adopted it on June 16 and the Council signed off on June 29. It amends the original 2024 AI Act to push back Chapter III’s high-risk system rules: obligations tied to Annex III use cases, including biometric identification, employment, credit and education, now apply from December 2, 2027, and obligations for AI systems embedded in regulated products such as machinery and medical devices move to August 2, 2028. Both dates land more than a year later than originally scheduled. The Omnibus’s own recitals give the Commission’s rationale: preparation of the technical standards providers need to demonstrate compliance had fallen behind schedule, leaving “a compliance burden that is heavier than expected.”
The disclosure duties in Article 50 were not part of that delay. Providers must still design AI systems so people know when they are talking to a chatbot, unless that is already obvious. Generative AI outputs need machine-readable markers, with an exception for tools performing only “assistive” editing. And deployers who use AI to generate or manipulate a deepfake depicting a real, identifiable person, or to publish AI-generated text on public-interest topics without an editor’s sign-off, must disclose that AI was involved, with a lighter version of the duty for satirical, fictional or artistic content. The law frames these duties in terms of “deployers” generally, not platforms specifically, though the primary text available does not spell out where that category’s boundaries sit for an individual creator using a third-party AI tool.
The Omnibus did add a grace period on the marking requirement, but only for systems already on the market before August 2, giving those providers until December 2, 2026, to comply. The deepfake and chatbot disclosure duties carry no such grace period.
How Advertisers and Rights Groups Read the Guidelines
The European Commission published guidelines on July 20 to help providers and deployers apply Article 50, alongside a voluntary Code of Practice the Commission and the EU’s AI Board have endorsed as an adequate compliance route. “With today’s guidelines, the Commission supports the smooth and effective application of the AI Act to make AI systems interacting with people, such as chatbots and AI agents and AI content, more transparent and trustworthy,” said Henna Virkkunen, the Commission’s EVP for Tech Sovereignty, Security and Democracy.
The European Advertising Standards Alliance, the self-regulatory body for European ad standards, welcomed the guidelines’ effort to define “deep fake” but said the Commission’s own guidance strayed from the law it interprets, “regretting a tendency in the Guidelines to extend the concept of ‘deep fake’ beyond the wording of the AI Act.” EASA separately welcomed the guidance’s recognition that some AI-generated ad content can qualify for a lighter disclosure regime as creative or satirical work.
A coalition including European Digital Rights (EDRi), Access Now, the European Center for Not-for-Profit Law (ECNL) and Amnesty International objected to the Omnibus itself rather than to the disclosure rules. The groups said the law “delays key protections, weakens transparency and creates a dangerous precedent,” pointing specifically to a change that lets providers upload less information to the EU’s public AI database when they classify their own systems as not high-risk, a change the groups argue limits regulators’ and researchers’ ability to check those self-classifications.
A Separate, Older Rule on Training Data
A less-discussed AI Act provision has already been in force for a year. Article 53 has required providers of general-purpose AI models, since August 2025, to publish a summary of the content used to train their models, maintain technical documentation, and put in place a policy for complying with EU copyright law, including honoring rights reservations set out under the EU’s 2019 copyright directive.
A coalition of European creative-industry associations, including the International Confederation of Societies of Authors and Composers (CISAC), the European Grouping of Societies of Authors and Composers (GESAC), the International Federation of the Phonographic Industry (IFPI) and News Media Europe, welcomed the provision back in March 2024, when the AI Act cleared Parliament, calling it a first step toward letting rightsholders enforce their rights against AI developers and pressing for direct involvement in setting the technical template for those training-data summaries. That statement predates this month’s disclosure deadline by more than two years, and no rightsholder claim or enforcement action tied to Article 53 has been reported since it took effect.
The guidelines meant to help providers meet the disclosure duties that did take effect on schedule arrived just 13 days before the deadline.
Subscribe to Our Newsletter
Check Out Our Podcast