AI Act Enters into Force: What Changes for Companies Starting August 2?

Starting August 2 2026, AI transparency rules are no longer just a future obligation, as they are now being actively applied.

Companies should now be able to clearly state when their customers are communicating with artificial intelligence, how published content is generated, and who is responsible for using the given AI system. This does not mean, however, that every image, text, or email created with AI assistance must automatically receive a visible label.

Article 50 of the AI Act contains several distinct obligations applying to different situations and actors. The decisive factor is not merely whether AI was used, but also who uses it, for what purpose, and who the output is intended for.

A chatbot must disclose that it is a chatbot

One of the most visible changes applies to chatbots, voicebots, avatars and other systems interacting directly with humans. Users must be clearly informed at the very latest during the first interaction that they are communicating with artificial intelligence.

It is not enough to hide this information in terms and conditions, technical documentation, or a general statement that the service may use AI. For a chatbot, this can be an introductory message or a visible label, while for a voicebot, a short audio notice at the beginning of the call suffices. The more humanlike a system appears, the more critical its unambiguous identification becomes.

Interestingly, the rules also cover AI agents. If an agent communicates with a customer, supplier, or another person, it should not only state that it is an artificial intelligence. It should also disclose who it is acting on behalf of. This applies, for example, to an automated sales assistant or customer support operating on behalf of a specific company.

Not all AI content needs a visible label

Labeling content is one of the most frequently oversimplified parts of the AI Act. In reality, it is necessary to distinguish between two levels:
‍

• Technical transparency: Providers of generative AI systems must ensure that their outputs are marked in a machine-readable format such as metadata or invisible watermarks. For systems placed on the market before August 2, 2026, a transition period applies until December 2, 2026.
  ‍
• User transparency: This applies primarily to deepfakes and texts informing the public on matters of public interest. Manual labeling of every standard output is not required.

When is an AI image considered a deepfake?

An image or video created using AI is not automatically a deepfake. It must consist of content that resembles a real person, place, object, or event, creating a false impression among the audience that it is an authentic recording of reality.

Czech Association for Artificial Intelligence provides a practical example:
‍

• Product visualization: If an e-commerce shop uses AI to place a photograph of a chair into a generated living room, it is not a deepfake, provided the customer understands the scene as a visualization and the product features remain accurate.
  ‍
• News coverage: If a news portal uses AI to add a larger crowd to a photo from a protest, this misleads an audience expecting a real event recording.

The key factors are context, audience expectations and the risk that synthetic content will be mistaken for reality.

What about texts on matters of public interest?

A special regime applies to AI-generated or significantly modified texts published to inform the public on matters of public interest such as politics, public administration, health, safety, the environment, or the economy.

Such text must be labeled as generated or modified by AI. An exception applies if it undergoes genuine human or editorial review and a specific person takes responsibility for its publication. Mere grammar checking is not enough, as a human must evaluate the content substantively. Standard marketing text, product descriptions, or creative writing generally do not fall under this obligation.

The role of the company: Provider vs. Deployer

The AI Act differentiates obligations based on role:

A company cannot simply claim that it uses an external model and that all responsibility lies with its operator. If a company has a custom AI product built under its own brand, its own obligations apply.
‍

Violations of transparency rules can lead to fines of up to 15 million euros or 3% of total worldwide annual turnover, assessed relative to company size and severity.

What should companies do now?

The first step is gaining an overview of where AI is used within your company. Consider the following questions:
‍

• Do you have a chatbot, voicebot, or AI agents on your website?
• Do you create realistic photos, videos, or synthetic voices?
• Do you publish AI-generated text regarding topics of public interest?
• Who checks each output and takes responsibility for it?

The goal is not to mechanically add an AI icon to every output, but rather to identify situations where a customer could be misled.

Transparency is not just a legal obligation

The AI Act raises key product questions: Do we know where we use AI in our business? Do we understand what data it processes? Can customers tell if they are talking to a human? Is it clear who is responsible for the result? If you cannot answer these questions, it may indicate that the AI solution was not sufficiently thought through during its initial design.

At Techmates, we do not view transparency as a sticker added right before launch. Security, data handling, user trust, and human responsibility should be part of the solution from the start. The AI Act does not merely remind companies what to label, it forces them to clarify where, why, and under whose responsibility they are using artificial intelligence.

Would you like to use AI in your company safely and effectively? Schedule a consultation with us and together we will find a reliable AI solution that fits both your business needs and current regulations.