Thought Leadership
Starting August 2, 2026, Article 50 of the EU AI Act will apply. Chatbots must identify themselves, while deepfakes require technical labeling. Violations can result in substantial fines.
Companies operating chatbots or publishing AI-generated content in the EU should have this coming Sunday marked on their calendars. On August 2, 2026, the transparency requirements set out in Article 50 of the European Union’s AI Act become legally binding. For many organizations, this marks the end of a multi-year transition period that began when the regulation entered into force in August 2024.
What Takes Effect Sunday
At its core, the rules are designed to make it clear to users when they are interacting with a machine. Chatbots and similar automated systems must disclose their nature unless this is already obvious to the user. Deepfakes and other AI-generated or manipulated image, audio, and video content must carry a technical identifier, such as machine-readable markers. AI-generated text covering matters of public interest must also be labeled as such unless it has been reviewed by a human who assumes editorial responsibility.
In Germany, enforcement falls under the responsibility of the Federal Network Agency, while the EU AI Office will take on expanded oversight powers at the European level. However, the AI Act does not provide a definitive technical specification for how labeling must be implemented. Standards are therefore likely to emerge over the coming months.
Digital association Bitkom has compiled the key rules that become binding on Sunday:
- The relevant authorities may monitor compliance with these transparency requirements and impose fines for violations. Starting Sunday, this will also apply to breaches of the requirements for general-purpose AI models that have already been in effect since 2025.
- Chatbots must identify themselves as such. Users interacting with an AI system should generally be made aware that they are communicating with a machine.
- Newly launched generative AI systems must use machine-readable labeling to identify artificially generated or modified content. Systems already in operation have a transition period until December 2026.
- Deepfakes, meaning highly realistic artificially generated or manipulated images, videos, and audio recordings, must be disclosed.
- AI-generated text on matters of public interest must be identified as such unless a human has reviewed the content and assumes editorial responsibility for it.
- People must be informed when emotion recognition or biometric categorization is being used.
Not Everything Is Happening as Originally Planned
Companies concerned that the entire AI Act will suddenly come into force on Sunday can breathe a little easier. The requirement for digital watermarks for providers of generative AI, originally scheduled for August 2, has been postponed until December 2026. Chatbots and generative AI systems already in operation will also have a transition period until December 2, 2026, to implement the necessary technical changes.
The delays are significantly more substantial for high-risk AI systems. Under the EU Digital Omnibus (Regulation 2026/1744), which entered into force on July 27, 2026, the main obligations for high-risk systems listed under Annex III have been pushed back to December 2027. For AI systems embedded in regulated products such as vehicles or medical devices under Annex I, the deadline has been extended to 2028. Companies that have been waiting for the full high-risk AI framework will therefore have to wait longer.
A Look at the Timeline
The AI Act has been coming into force in stages since it entered into force in August 2024. Since February 2025, the ban on certain AI practices and requirements for AI literacy within companies have applied. Since August 2025, the governance rules and requirements for providers of general-purpose AI (GPAI) models, including large language model providers, have been in effect.
August 2, 2026, now marks the most extensive step so far: the introduction of transparency requirements covering almost all providers and operators of AI systems that have direct interaction with users.
Labeling Is Only the Beginning
Simon Hayward, VP of Sales at software provider Freshworks, told it-daily.net that labeling is the most visible, but by no means the most important, part of a much broader transformation. In his view, companies face a much more fundamental question: Do they actually know where AI is already being used across their business processes?
A label alone does not explain how content was created, whether a human reviewed it, or who ultimately assumes responsibility for it. Hayward therefore argues that companies need to build governance structures and establish clear responsibilities from the outset rather than trying to retrofit compliance later.
His central argument is that speed and efficiency will no longer be enough to succeed with AI. Trust, he believes, will become the real competitive differentiator.
The more important question, therefore, is whether companies actually know where AI is being used across their business processes. As AI becomes increasingly integrated into customer communications, internal operations, HR, marketing, and service processes, transparency becomes more critical, and just as important as innovation itself.
True transparency, however, can only be achieved by taking a holistic view of service processes, operational workflows, and corporate resources, while building reliable AI applications on a solid data foundation.
Simon Hayward, VP of Sales at Freshworks
What Companies Should Do Now
Organizations using their own chatbots, image generators, or analytics tools, including those integrated into third-party software, should immediately assess whether and how users are being informed about the use of AI.
Violations of the transparency requirements can result in fines of up to €15 million or 3 percent of a company’s worldwide annual turnover, whichever is higher.
(lb)
