“Please open your doors for AI to make use of your data.”
That is not a vendor slogan. It is what Chancellor Friedrich Merz told an audience of company leaders in Berlin this week, sitting next to NVIDIA CEO Jensen Huang. He addressed the Mittelstand directly.
Some managers will hear it as an official all-clear: if the Chancellor says so, the team can finally use ChatGPT and Claude for real work. Others will hear the opposite and keep their ban in place.
Both readings miss what was actually said. So let us examine the evidence.
Information checked on 22 October 2026. Quotes are taken from the English-language conversation as recorded and published by NVIDIA and heise, supplemented by a transcript of the recording. This article is not legal advice.
TL;DR
- Yes, it is a public invitation to use AI. Merz explicitly asked the Mittelstand to use AI with its company data.
- No, it is not a recommendation for ChatGPT or Claude. Neither product was mentioned. The conversation was about industrial and physical AI.
- And it is not a blank cheque for your data. Huang argued that Germany should not export its knowledge and import its own intelligence back.
- Your employees are probably not waiting for permission anyway. The real question is under whose rules AI is used.
What was actually said in Berlin
The conversation opened the two-day Invest in Germany summit at Kraftwerk Berlin. It came alongside a joint declaration of intent between the German government and NVIDIA on physical AI and the Industry 4.0 platform.
Merz called the Mittelstand “the biggest advantage we have in Germany” and continued:
“Please, open your doors for AI to make use of your data. This is so decisive for the German future.”
Huang picked up the thought, with a twist many headlines skipped:
“The data of Germany is part of your natural resources. It belongs to the nation. […] You should import the technology to harvest that data and turn it into your national intelligence, but you should not export your culture, export your values, export your knowledge, export your history, and then import German intelligence back.”
A few sentences later, he spoke directly to the Mittelstand. Experienced staff are retiring, and their craft goes with them:
“Take advantage of artificial intelligence. Let them codify. Let them program. Let them teach this artificial intelligence and keep it here in Germany.”
The evidence check: three claims, three verdicts
Claim 1: “The government wants companies to use AI.”
Verdict: confirmed. There is little ambiguity. The Chancellor did not just tolerate AI; he asked companies to use their own data with it. If your AI policy is still “we will wait and see”, you can no longer claim to be following the political mood.
Claim 2: “That means: use ChatGPT and Claude.”
Verdict: not stated, but not ruled out. Neither name appeared. The examples were autonomous cars, robotics, digital twins and factories, not drafting emails.
Huang did not argue for isolation either: companies would “of course use the best AI technology available in the world and AI services in the world, but you have every capability to build your own AI.” Using leading international models is therefore part of the picture. They are tools, not the place where your knowledge should end up.
Claim 3: “So we can upload our company data.”
Verdict: refuted as a blanket statement. The central idea in Huang’s answer is the exact opposite of “paste everything into a chat window”. Your process knowledge, product data and the experience of your senior employees are described as a resource. You import the technology to use them. You do not give them away.
Overall verdict: It is a public invitation to use AI. It is not a permission slip for uncontrolled use of ChatGPT, Claude or any other service with company data.
The uncomfortable part: the doors are already open
Many companies discuss whether to allow ChatGPT while their employees have long since decided for themselves.
In a Bitkom survey of 604 companies with 20 or more employees, published in October 2025, 8% said private AI use at work was widespread and another 17% reported individual cases. Only 26% provided employees with official access to generative AI. Only 23% had rules for its use.
That is the provocation the Chancellor’s sentence does not resolve: a ban that nobody enforces is not a data protection strategy. It simply moves AI usage to private accounts, outside your contracts and outside your view.
What actually differs between a private account and a business setup
The same model brand can come with very different terms. Before you decide, check at least these points:
- Training on your data: OpenAI states that it does not train on data from ChatGPT business plans or its API by default. Anthropic states the same for inputs and outputs from its commercial products. Exceptions such as opt-ins or submitted feedback should be part of your policy.
- Where data is stored and processed: OpenAI offers European data residency for new ChatGPT Enterprise and Edu workspaces and new API projects. Processing locations differ by provider, product and contract. Check them for the plan you actually buy rather than relying on a vendor headline.
- Contract and GDPR: You need a data processing agreement, a legal basis for personal data and a decision about which data may be used at all. A business plan does not replace these steps.
- AI literacy: Article 4 of the EU AI Act requires providers and deployers to take measures to support the AI literacy of people using AI systems on their behalf. Handing out accounts without guidance does not meet that intention.
A pragmatic traffic light for your data
We find a simple three-level classification more useful than a general yes or no:
- Green: public or uncritical content. Website copy, generic research, wording help, learning. An approved business account is sufficient.
- Amber: internal but not personal. Process descriptions, internal documentation, quotation templates. Only in approved tools with a contract, no training on your data and clear rules.
- Red: personal data, trade secrets, production and design knowledge. Customer and employee data, calculations, design data, the know-how of your best engineers. Only with a deliberately designed architecture: defined data locations, permissions, logging and a model connection you have checked.
The red category is what Huang meant by national resource. For a Mittelstand company it is simply the core of its competitiveness.
Five steps you can take this month
- Ask instead of assuming. Find out anonymously which AI tools your teams already use and for which tasks.
- Provide an approved option. People use private accounts when there is no official alternative.
- Publish a one-page rule. The traffic light above, plus examples from your own company, is enough to start.
- Train with real tasks. Show what AI does well, where it makes things up and what must never go into a prompt.
- Choose one red-zone use case deliberately. For example, capturing the knowledge of an experienced service technician in a knowledge system that stays under your control.
The last step is where Huang’s argument becomes concrete. Your knowledge base does not have to live in a chat history at a foreign provider. With LIVOI, the application and central data stores run in a German data center. Model calls and external services are configured and assessed separately. That separation is exactly the decision every company should make consciously, whichever platform it uses.
When this article does not help you
If your company processes particularly sensitive data, such as health data or classified information, you need a specific legal and security assessment. A traffic light is a starting point, not a substitute for a data protection impact assessment.
Common questions
Did Merz recommend ChatGPT or Claude?
No. Neither product was mentioned. He called on companies to use AI with their data. The examples in the conversation came from industrial and physical AI.
Is it illegal to use ChatGPT or Claude at work in Germany?
No. But the use of personal data requires a legal basis, a suitable contract and appropriate safeguards. Pasting customer data into a private account is a different matter from using a contracted business plan.
Is a private account acceptable for work if I am careful?
We advise against it. You lose control over contractual terms, retention and settings. An approved business option is usually cheaper than the risk.
Does “keep it here in Germany” mean only German AI models?
Not according to the conversation. Huang expected companies to use the best AI services available. The point is control over your own data and knowledge.
Not sure where to begin? We help companies find use cases that create value without giving away their knowledge.