AI Act and transparency: what must companies show from August?
August 2026 isn't just about lawyers and high-risk systems.
For many companies, the most concrete question is simpler:
Does the customer know when they're interacting with AI?
AI Act Article 50 sets transparency requirements for certain AI systems. It affects chatbots, AI-generated content, deepfakes, and systems that manipulate image, audio, video, or text.
Start with the customer-facing surfaces
List every place where a person can encounter AI from your company.
Examples:
- chatbot on the website
- AI assistant in the customer portal
- automatic support suggestions sent to customers
- AI-generated product images
- AI voice in sales or support flows
- AI-generated articles, guides, or news
- synthetic images or videos of people
Then mark three things:
- Does the user realize AI is involved?
- Does it happen at first interaction or much later?
- Is there human editorial control where the content informs the public?
If the answer is unclear, you need to change the flow.
The chatbot shouldn't pretend to be human
If an AI chatbot interacts directly with a person, the person must be informed that they are communicating with an AI system — unless it's obvious from the context.
Write briefly and clearly.
Example:
"You're chatting with an AI assistant. It can help with common questions. A human takes over for cases that require decisions or personal data."
That's better than a vague line in the privacy policy.
AI-generated content needs accountability
Article 50 also targets synthetic and manipulated content.
For a company publishing content, that means a practical question:
Is the text, image, video, or voice created or manipulated by AI in a way that the recipient needs to understand?
You don't need to label all internal drafts. But you need a policy for public content.
Especially if the content:
- depicts real people
- could be mistaken for real image, audio, or video
- informs the public on matters of public interest
- lacks human editorial control
Build transparency into the content workflow
A good content workflow should log:
- which tool created the text or image
- which prompt or brief was used
- who edited
- who approved
- whether AI disclosure is required
- where disclosure is displayed
It sounds like a lot. In practice, it can be a few columns in a spreadsheet.
The important thing is that the decision isn't made ad hoc every time someone publishes.
Three things to do now
1. Add disclosure to the AI chat
Show it at first interaction. Write in plain language.
2. Create an AI content policy
Decide when AI-generated material should be labeled, when a human editor takes responsibility, and what you don't publish.
3. Connect labeling to approval
No publication of AI-generated image, video, voice, or informational content without someone checking the transparency requirement.
The readiness question
If someone asks in August: "How do your customers know when they're interacting with AI?"
You should be able to answer with a workflow, not a feeling.