The EU started asking OpenAI and Anthropic for information on 14 October

The Commission has started asking questions

On 14 October, the European Commission asked OpenAI and Anthropic for information under Article 90 of the AI Act ^[1]. Article 90 gives the Commission the power to request all necessary documents and data from providers of general-purpose AI models, and to conduct interviews ^[2].

The requests have not been published in full, but according to reports they focus on the transparency obligations in Article 50 and the data governance requirements in Article 53 ^[1]. Both articles apply to general-purpose AI models, including the large language models that power ChatGPT and Claude.

This is not an enforcement action. It is the Commission using its supervision powers to check whether providers are meeting their obligations. But it is one of the first public signs that supervision of generative AI under the AI Act has begun in practice.

What the rules say

Article 50 requires providers to tell people when they are interacting with an AI system, and to mark AI-generated content in a machine-readable way ^[3]. It also asks deployers to disclose when they publish AI-generated text to inform the public, unless a person has reviewed it and taken responsibility for it. Article 50 has applied since 2 August 2026.

Article 53 sets out data governance obligations for providers of general-purpose AI models. They must put in place a policy to respect Union copyright law, publish a sufficiently detailed summary of the content used for training, and comply with a request to put the policy into practice through state-of-the-art technologies ^[4]. Article 53 came into force on 2 August 2024, earlier than most of the AI Act.

Think of it like asking a bakery for the recipe. The bakery can keep the exact proportions to itself, but it has to list the ingredients and show it isn't using anything it shouldn't. Article 53 works the same way: the model's weights stay private, but the training data policy and the summary have to be public.

Article 90 gives the Commission the tools to check. It can ask for documents, data and explanations, conduct interviews and, if a provider does not cooperate, carry out inspections ^[2]. The Commission can also ask a provider to take measures to comply with the AI Act, and if the provider does not do so within a set period, the Commission can impose periodic penalty payments or fines.

What this means for you

If you provide or deploy a general-purpose AI model, this is worth noting:

• The Commission is using its supervision powers under Article 90. Requests for information are not sanctions, but they are a step towards checking compliance.
• Article 50 and Article 53 are in force. If you have not yet published a training data summary or put a copyright policy in place, that is a gap the Commission can ask about.
• Transparency obligations go beyond the model card. Article 50 asks for machine-readable marking of generated content, and deployers of systems like chatbots must tell users they are talking to an AI.
• The evidence question is whether you can show what you told users and when. If the Commission or a user questions what your AI said or disclosed, your own record is easier to check than theirs.

A digital witness for what your AI disclosed

This is a case where a digital witness fits well. The thing to seal is each disclosure made under Article 50: the notice to the user that they are interacting with an AI, or the record of what content was marked as AI-generated and when.

Sealing takes a fingerprint of the record, not the record itself, so the disclosure stays in your systems. When a user or a supervisor questions what was disclosed, anyone you share the record with can check that it is exactly what was written at that time.

A digital witness does not show that the disclosure was adequate or that the model complies with Article 53. It shows that a record of the disclosure existed and has not changed since it was sealed.

You seal the record when the disclosure is made, through the REST API, the SDK or the MCP server. Your system sends the fingerprint to WitniumChain, gets a witness ID back and stores it with the record. To check, you or the person questioning it calculates the fingerprint again and compares it with the sealed one. If it matches, nothing has changed. Change a single character and it will not match. The content never leaves your systems, and the first 2,000 witnesses a month are free. Read the full technical documentation at witniumchain.com/docs.

A digital witness does not replace the disclosure duty. It gives you a checkable record of having met it.

How it would work

1. Your chatbot tells a user they are talking to an AI. Your system writes a record of the disclosure (user session reference, timestamp and the text shown).
2. At the same time, it sends a fingerprint of that record to WitniumChain and gets a witness ID back.
3. Six months later, the user or a supervisor asks what the chatbot disclosed. You share the record and the witness ID. They calculate the fingerprint and check it against the sealed one. If it matches, the record has not changed since the disclosure was made.

What to watch

The next step to watch is whether the Commission publishes more detail about what it asked for, and whether it takes further action under Article 90. Providers of general-purpose AI models must also watch for guidance from the AI Office on what counts as a sufficiently detailed summary under Article 53.

For deployers, the transparency deadline under Article 50 is already here. If your AI talks to customers or generates content for the public, the question is whether you have a record of what it disclosed, and whether that record would hold up if someone questioned it.