Artificial intelligence is already an integral part of numerous business processes. It is used to handle enquiries, generate content, analyse information, automate tasks and support certain decisions.<\/p>\n
From 2 August 2026<\/strong>, the transparency obligations set out in Article 50 of Regulation (EU) 2024\/1689 on artificial intelligence, known as the AI Act<\/strong>, apply across the European Union.<\/p>\n These obligations are intended to ensure that people are aware when they are interacting directly with certain AI systems and can recognise content that has been artificially generated or manipulated.<\/p>\n However, this does not mean that every use of artificial intelligence must be public, nor that all content created with the aid of AI must carry a visible label. The obligations depend on the type of system, the role played by the company and the specific use of the technology.<\/strong><\/p>\n For organisations, the question is no longer simply whether they use artificial intelligence. They must also identify where they use it, for what purpose, who is involved in the process and what measures of transparency, oversight and control apply<\/strong>.<\/p>\n Article 50 of the AI Act sets out specific obligations for certain providers and deployers of artificial intelligence systems<\/strong>.<\/p>\n Among other scenarios, it applies to systems that interact directly with people, tools that generate synthetic content, emotion recognition or biometric categorisation systems, and certain content generated or manipulated using AI.<\/p>\n For many companies, this means moving from an approach to AI based mainly on individual or departmental decisions to a more structured management of the tools, their uses and their risks<\/strong>.<\/p>\n Before analysing the transparency obligations, it is important to bear in mind a very recent amendment to the European framework.<\/p>\n The Digital Omnibus on AI<\/strong>, adopted as Regulation (EU) 2026\/1744, introduced specific adjustments to the AI Act with the aim of simplifying its application. These amendments came into force on 27 July 2026<\/strong>, just a few days before the key date of 2 August.<\/p>\n One of its main effects has been to amend the timetable for certain obligations applicable to high-risk systems. The regulations for high-risk systems used in areas such as employment, education, biometrics, critical infrastructure and migration will apply now from 2 December 2027<\/strong>, whilst certain high-risk systems embedded in regulated products have until 2 August 2028 <\/strong>to comply.<\/p>\n However, this postponement does not generally apply to the transparency obligations set out in Article 50<\/strong>, which remain applicable from 2 August 2026.<\/p>\n The Omnibus does, however, introduce a specific provision for AI systems that were already on the market before 2 August 2026 and are subject to the labelling requirement set out in Article 50(2). Their providers have until 2 December 2026<\/strong> to take the necessary measures to comply with this requirement.<\/p>\n The new European timetable should therefore not be understood as a general postponement of the AI Act. Each obligation has its own application date.<\/p>\n The Regulation does not apply solely to companies that develop artificial intelligence models or tools.<\/p>\n The regulation distinguishes between several stakeholders. These include providers<\/strong>, who develop an AI system or commission its development and market it or put it into service under their own name or brand, and deployers<\/strong>, who use AI systems under their authority in the course of a professional activity.<\/p>\n This means that the regulation\u2019s scope for businesses is much broader than it might appear.<\/p>\n A company may not develop its own technology and may still use artificial intelligence systems to serve customers, create content, analyse information, support internal processes or incorporate certain functionalities into its products and services.<\/p>\n Providers established outside the European Union may also fall within the scope of the AI Act where the territorial conditions set out in the Regulation are met.<\/p>\n Therefore, to determine an organisation\u2019s obligations, it is not enough simply to identify who developed the tool. It is necessary to analyse what role the company plays and how it actually uses the AI system<\/strong>.<\/p>\n One of the easiest obligations to identify concerns systems designed to interact directly with people.<\/p>\n In such cases, the provider must design and develop the system in such a way that the person is aware they are interacting with artificial intelligence, unless this is already obvious from the circumstances and context.<\/p>\n The information must be provided from the start of the first interaction, in a clear and distinguishable manner, and in accordance with the relevant accessibility requirements<\/strong>.<\/p>\n This obligation may apply, for example, to certain chatbots, AI agents, avatars or conversational systems<\/strong> used in customer service.<\/p>\n Companies incorporating these types of solutions should therefore check how the tool is presented to the user and whether the automated nature of the interaction is made sufficiently clear.<\/p>\n No.<\/p>\n This is one of the issues that raises the most questions regarding the application of Article 50.<\/p>\n The Regulation states that providers of systems that generate synthetic text, image, audio or video<\/strong> content must take measures to ensure that such outputs can be identified as generated or manipulated by artificial intelligence.<\/p>\n This obligation generally consists of incorporating a machine-readable marking<\/strong> and enabling the content to be detected as artificially generated or manipulated.<\/p>\n There are, however, exceptions and nuances. The European framework itself provides for certain auxiliary uses or modifications that do not substantially alter the content or its meaning.<\/p>\n It is important to distinguish this technical obligation, which applies to certain providers, from the obligation to display a visible or perceptible notice to people<\/strong>.<\/p>\n Not all AI-generated content requires a visible label. The AI Act reserves this obligation for specific cases.<\/p>\n One such example is deepfakes<\/strong>.<\/p>\n The AI Act defines a deepfake as any image, audio or video content generated or manipulated using artificial intelligence that resembles existing people, objects, places, entities or events and could falsely appear to be authentic or truthful.<\/p>\n When a company or professional uses an AI system to generate or manipulate this type of content and disseminates it as part of their professional activity, they must clearly and conspicuously disclose its artificial nature.<\/p>\n For works that are clearly artistic, creative, satirical, fictional or similar, the Regulation allows this obligation to be met in a proportionate manner that does not unnecessarily interfere with the viewing or enjoyment of the work.<\/p>\n Context is essential. Not every image or video retouched using AI automatically constitutes a deepfake<\/strong>. It is necessary to analyse, among other factors, whether there is a false appearance of authenticity or truthfulness.<\/p>\n Texts generated or manipulated using artificial intelligence may also be subject to a visible transparency obligation when published for the purpose of informing the public about matters of public interest<\/strong>.<\/p>\n This may include issues relating to politics and democratic processes, public services, justice, fundamental rights, security, public health, environmental protection or certain economic, financial, scientific and cultural events that are relevant to public debate.<\/p>\n However, there is a particularly important exception for companies, media and communications departments.<\/p>\n If the text has undergone genuine human review or editorial scrutiny, and a natural or legal person assumes editorial responsibility for its publication, it is not mandatory to label it on this basis as text generated or manipulated using AI.<\/strong><\/p>\n The review must be substantive. A purely formal check, such as correcting spelling or grammar mistakes, is not in itself sufficient human review. The Commission requires genuine oversight and for someone to assume editorial responsibility for the publication.<\/p>\n For companies using artificial intelligence in marketing, communications or content production, this point reinforces the importance of establishing clear processes for review, validation and editorial responsibility<\/strong>.<\/p>\n Article 50 also sets out obligations for deployers of certain emotion recognition and biometric categorisation systems<\/strong>.<\/p>\n Where such systems are used within the scope permitted by the regulations, the people exposed to them must be informed about the use of these technologies.<\/p>\n It is important to bear in mind that complying with a transparency obligation does not render lawful a use that may be prohibited under other provisions of the AI Act.<\/p>\n These are also particularly sensitive technologies from the perspective of fundamental rights and privacy. Organisations using them must therefore also assess their compliance with data protection legislation<\/strong> and with any other regulations applicable to the specific case.<\/p>\n The entry into application of these obligations does not mean that all companies need to change all their processes or indiscriminately label every use of artificial intelligence.<\/p>\n It is, however, advisable to know precisely which AI systems the organisation actually uses and the role each one plays<\/strong>.<\/p>\n A first step may be to draw up an inventory of the tools used in areas such as marketing, customer service, human resources, sales, administration, data analysis or product and service development.<\/p>\n The next step is to determine what role the company plays in each case, which people may be affected and what type of outputs the tool produces.<\/p>\n Among other issues, organisations should review:<\/p>\n The aim is not to apply labels automatically to every item of content, but to establish controls that are proportionate to the specific use of artificial intelligence<\/strong>.<\/p>\n Transparency towards users is only one part of proper corporate governance of artificial intelligence.<\/p>\n Organisations should also establish internal measures to know which tools are authorised, who can use them, what information may be entered, how outputs are reviewed and how to respond to a potential incident<\/strong>.<\/p>\n The AI literacy<\/strong> requirement set out in Article 4 of the AI Act also remains relevant.<\/p>\n The Digital Omnibus has amended this obligation. Providers and deployers are still required to take measures to support the development of AI knowledge and skills among people working with these systems, but the Regulation no longer requires them to guarantee a specific or \u201csufficient\u201d level of literacy for each individual.<\/p>\n The approach must be adapted to the context, the knowledge, experience, education and training of the people involved, and the systems actually used by the organisation.<\/p>\n Transparency, training, documentation and supervision should therefore not be treated as separate elements. They form part of the same artificial intelligence governance strategy.<\/p>\n Although the AI Act applies generally from August 2026, the Regulation sets out a phased timetable<\/strong>, and the Digital Omnibus has amended some of its dates.<\/p>\n Prohibitions on certain practices and the AI literacy obligations came into force in February 2025. Governance rules and certain obligations relating to general-purpose AI models came into force in August 2025. The transparency obligations under Article 50 apply from 2 August 2026.<\/p>\n Following the Digital Omnibus, certain regulations relating to high-risk systems are set to apply from 2 December 2027<\/strong> for high-risk systems listed in Annex III and from 2 August 2028<\/strong> for high-risk systems embedded in regulated products.<\/p>\n Each company must therefore analyse which obligations are already applicable to its systems and which require preparation for later phases.<\/p>\n Compliance with Article 50 falls mainly to the competent national market surveillance authorities<\/strong>.<\/p>\n The European AI Office has powers in certain cases, whilst the European Data Protection Supervisor performs the relevant functions in relation to systems used by European Union institutions, bodies and agencies.<\/p>\n For certain infringements, the AI Act provides for fines of up to \u20ac15 million or 3% of the annual global turnover for the previous financial year<\/strong>, with proportionality criteria for SMEs and small mid-cap companies.<\/p>\n Beyond the potential sanctions\u2019 regime, the significant change for organisations is that certain transparency measures are no longer dependent solely on each company\u2019s best practices. They are becoming part of the applicable regulatory framework.<\/p>\n Only in one specific case.<\/p>\n There is a transition period until 2 December 2026<\/strong> for AI systems that were already on the market before 2 August 2026, and only in relation to the technical obligation to label and detect AI-generated content under Article 50(2)<\/strong>.<\/p>\n There is therefore no general grace period until December for all transparency obligations under Article 50.<\/p>\n In addition, content generated before 2 August 2026 does not need to be labelled retroactively, although the Commission encourages doing so where possible.<\/p>\n Alongside the new obligations, the European Union has a Code of Practice on transparency for AI-generated content<\/strong>.<\/p>\n This is a voluntary instrument aimed at providers and deployers of generative AI systems. It offers practical measures to address the marking, detection and labelling obligations set out in Article 50.<\/p>\n Adherence to the Code is voluntary, but the legal obligations under the AI Act are not. The Commission and the AI Board have considered the Code an appropriate tool to demonstrate compliance with the obligations it covers. Organisations that choose other routes must be able to show that the measures they have taken are also appropriate.<\/p>\n The European Union has also developed icons to identify certain content generated using artificial intelligence<\/strong>. Their use is voluntary and does not in itself prove compliance with the Regulation.<\/p>\n Not generally speaking. The Digital Omnibus has amended several AI Act dates, particularly those relating to certain high-risk systems, but the obligations under Article 50 continue to apply from 2 August 2026<\/strong>.<\/p>\n Does content always have to be labelled as having been created using artificial intelligence?<\/strong><\/p>\n No. It depends on the type of content, the way in which artificial intelligence has been used and the role played by the company.<\/p>\n The Regulation distinguishes, for example, between the technical marking required of certain providers and the perceptible information obligations that apply to deepfakes or to certain texts on matters of public interest.<\/p>\n Not necessarily.<\/p>\n For texts published to inform the public about matters of public interest, labelling is not mandatory on this basis where there is genuine human review or editorial control<\/strong> and a natural or legal person assumes editorial responsibility for the publication.<\/p>\n A mere spelling or grammar check is not enough to benefit from this exception.<\/p>\n No.<\/p>\n They may also apply to organisations that use systems developed by third parties in the course of their professional activity.<\/p>\n That is why one of the first issues a company must analyse is what role it plays in relation to each system and what its specific obligations are.<\/p>\n Where the conditions set out in Article 50 for systems that interact directly with people are met, the user must know from the start of the interaction that they are dealing with an artificial intelligence system, unless this is evident from the context.<\/p>\n There is a limited transition period until 2 December 2026<\/strong> for the technical obligation regarding labelling and detection set out in Article 50(2) in relation to systems that were already on the market before 2 August 2026.<\/p>\n This is not a general transition period for all Article 50 obligations.<\/p>\n In addition, content generated before 2 August 2026 does not need to be labelled retroactively.<\/p>\n At the same time, the recent Digital Omnibus has adjusted part of the AI Act timetable and certain obligations, making it even more important to analyse each case on an individual basis.<\/p>\n The underlying issue is broader. Organisations need to know which artificial intelligence they are using, for what purpose, which people are involved in the processes and what mechanisms exist to control the associated risks<\/strong>.<\/p>\n Companies that organise these procedures now will be better placed to comply with the obligations already in force and prepare for the next phases of the AI Act.<\/p>\n At Certus Legal Firm<\/strong>, we advise companies and organisations on digital law, data protection, new technologies and regulatory compliance. We analyse the use of artificial intelligence systems, the obligations that may apply and the measures needed to adapt processes, policies and procedures to the European and Spanish regulatory framework.<\/p>\n","protected":false},"excerpt":{"rendered":" Artificial intelligence is already an integral part of numerous business […]<\/p>\n","protected":false},"author":2,"featured_media":16673,"comment_status":"open","ping_status":"closed","sticky":false,"template":"","format":"standard","meta":{"footnotes":""},"categories":[225,64],"tags":[471,476,474,472,475,473],"class_list":["post-16688","post","type-post","status-publish","format-standard","has-post-thumbnail","hentry","category-new-technologies","category-legal","tag-ai-act-2-august-2026","tag-ai-act-for-businesses","tag-ai-generated-content-labelling","tag-article-50-ai-act","tag-artificial-intelligence-deepfakes","tag-artificial-intelligence-transparency"],"yoast_head":"\nWhat changes from 2 August 2026?<\/strong><\/h2>\n
What has changed with the Digital Omnibus on AI?<\/strong><\/h2>\n
Which companies are affected by the AI Act transparency obligations?<\/strong><\/h2>\n
When a person interacts directly with AI<\/strong><\/h2>\n
Must all AI-generated content be labelled?<\/strong><\/h2>\n
Deepfakes and content that appears to be real<\/strong><\/h2>\n
AI-generated texts on matters of public interest<\/strong><\/h2>\n
Emotion recognition and biometric categorisation<\/strong><\/h2>\n
What companies using AI should review now<\/strong><\/h2>\n
\n
Transparency, supervision and internal control of AI<\/strong><\/h2>\n
2 August does not mean that all AI Act obligations already apply<\/strong><\/h2>\n
Who oversees transparency obligations?<\/strong><\/h2>\n
Is there a transition period until December 2026?<\/strong><\/h2>\n
The Code of Practice can help demonstrate compliance<\/strong><\/h2>\n
Frequently asked questions about AI transparency obligations<\/strong><\/h2>\n
Has the Digital Omnibus delayed these obligations?<\/strong><\/h3>\n
Does an article written with the help of AI have to be labelled?<\/strong><\/h3>\n
Do the obligations apply only to technology companies?<\/strong><\/h3>\n
Is it necessary to inform users that a chatbot uses AI?<\/strong><\/h3>\n
Is there a transition period?<\/strong><\/h3>\n
From using artificial intelligence to governing it<\/strong><\/h2>\n