Meeting AI Transparency Obligations (Chatbots, AI Content, Deepfakes)

Does your company use an AI chatbot in customer service, or have texts, images, or videos created by artificial intelligence? Then a new labeling obligation applies from August 2, 2026, one that many companies have not yet had on their radar.

In short: Article 50 AI Act requires providers and deployers of generative AI, from August 2, 2026, to disclose an AI interaction, for example with a chatbot, and to label AI-generated or manipulated content as well as synthetic media. All companies that use AI in customer contact or for content creation are affected. Violations can result in fines of up to €15 million or 3% of worldwide annual turnover.

Who is affected by the AI transparency obligation?

This affects providers and deployers of generative AI systems used in business operations. That ranges from a chatbot on the company website or in customer service to AI tools that generate text, images, or videos for marketing, social media, or editorial purposes. Even companies that use AI tools only occasionally, for instance for individual ads, product images, or blog articles, fall under the obligation as soon as this content is published. Anyone who uses AI in customer contact or publishes AI-generated content must disclose the use, or label the content accordingly.

What exactly do you need to do?

The obligation under Article 50 AI Act comprises two separate requirements: disclosing AI interactions to users, and visibly labeling AI-generated or manipulated content. Both requirements apply independently of each other: a company can already be affected simply by using a chatbot, even if it publishes no AI-generated content, and vice versa. Implementing both requirements takes the following steps:

1. Record where AI is used in external-facing contexts (chatbots, text generation, image/video generation).
2. Design chatbots and AI assistants so that users can recognize the AI interaction.
3. Visibly label AI-generated or substantially altered content and synthetic media.
4. Prepare machine-readable marking of AI-generated outputs (grace period until December 2, 2026).
5. Document labeling rules internally and train editorial and marketing teams.

A tool such as Company Audit shows exactly which of these obligations apply to your company, letting you systematically match your own AI use against the legal requirements instead of checking every requirement one by one. It also makes sense to treat the five steps not in isolation but as a fixed sequence: first the inventory, then the technical implementation for chatbot and content, then the internal documentation and training. That way the obligation stays on the radar for future AI projects in the company too. The internal documentation and training of editorial and marketing teams in particular is often underestimated, but it is necessary so that labeling happens not just once but consistently for every new piece of AI content.

By when do you need to act?

The disclosure obligation for AI interactions and the visible labeling of AI-generated content apply from August 2, 2026. For the machine-readable marking, meaning the technical labeling of AI-generated outputs that systems can read, an additional grace period applies until December 2, 2026. Companies should therefore implement the visible labeling first and prepare the machine-readable marking in parallel, rather than tackling both at once shortly before the respective deadline. Unlike some individual high-risk requirements of the AI Act, the August 2, 2026 deadline for the transparency obligation has not been postponed so far; anyone counting on a delay here is taking an unnecessary risk.

What are the consequences of a violation?

Violations of the transparency obligations under Article 50 AI Act can be penalized with fines of up to €15 million or up to 3% of worldwide annual turnover. For smaller companies and sole proprietors, besides the financial risk, the loss of trust is particularly relevant: if it is not recognizable that a chatbot or a published piece of content comes from an AI, that can damage relationships with customers and business partners.

Frequently asked questions

Who must disclose AI interactions?

Providers and deployers of generative AI who use AI in customer contact, for example through a chatbot on the website or in customer service, must disclose from August 2, 2026 that users are interacting with an AI and not with a human.

Which content must be labeled?

Labeling is required for AI-generated or substantially altered content as well as synthetic media, for example AI-created texts, images, or videos that are published, regardless of whether they are used for marketing, social media, or editorial purposes.

What does the grace period until December 2, 2026 mean?

For the machine-readable, technical marking of AI-generated outputs, the legislator grants a transition period until December 2, 2026, in addition to the visible labeling. The visible labeling itself, by contrast, must already be in place from August 2, 2026.

How high are the fines for violations?

Violations of the transparency obligation can be penalized with fines of up to €15 million or 3% of worldwide annual turnover. This applies both to the failure to disclose AI interactions and to the failure to label AI content.

Is a simple notice text enough as labeling?

A visible notice such as "AI-generated" combined with a metadata marking covers the requirements twice over and is recommended, since it satisfies both the visible and the machine-readable labeling obligation.

Source: Article 50 AI Act. This article is general information and does not replace individual legal advice. As of: July 2026.