The proposal is supposed to be finished tomorrow. The client documents are extensive, and my AI account is at my fingertips. So I upload meeting notes, calculations, and a few internal documents. ChatGPT or Claude provide a usable draft within a few minutes.
This is a fictional example, but it’s a perfectly plausible situation in everyday consulting practice. AI can make research, structuring, and conceptual work easier for consultants. However, as soon as information from a client engagement is processed, another question arises: Is the consultant even allowed to share this data with the AI service being used?
The answer does not depend solely on the tool. The decisive factors are the data, the specific account, the terms of the agreement, and the agreed-upon permissions.
Effective Date: October 12, 2026.
Not across the board. Before entering any data, it is particularly important to clarify the contractual admissibility, the protection of confidential information, and the data protection requirements. The product version and settings of the AI service must also be reviewed.
A paid personal account is not automatically considered an approved business solution. Conversely, not all professional use of a personal account is prohibited. A general organizational matter that does not involve customer information should be evaluated differently than the upload of a confidential personnel plan.
As a general rule of thumb, confidential or personal customer data should not be shared with external AI services without prior review and the necessary approvals.
A consulting project can involve a wide variety of information: contact persons, interview notes, performance evaluations, prices, margins, sales strategies, or planned reorganizations.
Personal data is subject to the GDPR. A valid legal basis must exist for its processing and, where applicable, its transfer to an AI provider. The Data Protection Conference explains these requirements in its guidance on artificial intelligence and data protection.
Other information may not relate to specific individuals but may still be confidential. The fact that an internal price calculation does not contain any names does not make it freely usable. If a piece of information meets the legal requirements for a trade secret, it is additionally protected under the Trade Secrets Act. This includes, in particular, economic value and appropriate confidentiality measures. Section 2 of the Trade Secrets Act (GeschGehG) is decisive.
Data protection and confidentiality are therefore two separate assessments. One does not replace the other.
The fact that a customer provides documents means, first and foremost, that they may be used for the agreed-upon project. Whether external AI services can be used for this purpose is a separate question.
Consultants should, in particular, review confidentiality agreements, guidelines on the use of external service providers, and project-specific AI rules. A duty of confidentiality may preclude disclosure. With regard to trade secrets, Section 4 of the Trade Secrets Act (GeschGehG) contains corresponding prohibitions on certain actions.
Contractual claims for damages may also apply. However, they do not automatically arise from every erroneous prompt. Among other things, a breach of duty, liability for that breach, and resulting damage are decisive factors. This is provided for in § 280 of the German Civil Code (BGB).
The customer submitted some documents. A general power of attorney for each upload was not necessarily included.
"I have disabled training" is an important piece of information. It is not a complete assessment.
For ChatGPT Business and Enterprise, OpenAI states that it does not use business data for model training by default. However, the personal and business plans are distinct products with different security and management features, as shown in OpenAI’s official product information.
Anthropic also distinguishes between consumer offerings—such as Claude Free, Pro, and Max—and commercial offerings. For consumer offerings, Anthropic outlines several conditions under which chats may be used to improve models, including settings and explicitly provided feedback. The applicable data usage terms are decisive.
Regardless of the training, the service processes the input data to handle the request. Storage, access, deletion, integrated services, and potential transfers to third countries still need to be reviewed.
Therefore, the absence of model training does not mean either “no data processing” or “authorized by the customer.” Even a company’s offer does not replace an assessment of the specific use case.
A consultant may be a contractor under a contract without acting as a data processor under data protection law. The terms describe different relationships.
For the purposes of data protection law, the decisive factor is who determines the purposes and essential means of the respective processing. Depending on the activity and the actual level of involvement, different roles may apply. A contract title alone is not sufficient to determine this. The European Data Protection Board’s guidelines on data controllers and data processors explain this distinction.
If the consultant acts as a data processor and the AI provider is engaged as an additional data processor, particular attention must be paid to the client’s instructions, the required prior written authorization, and the contractual safeguards. Such authorization may be granted on a case-by-case basis or generally. The legal basis is Article 28 of the GDPR.
Furthermore, the customer's consent does not replace the necessary legal basis for processing the personal data of employees or business partners.
Not necessarily. A person’s position, location, project description, and specific events can make them identifiable even without mentioning their name. The Data Protection Conference explicitly points this out.
Pseudonymized data is, by definition, still personal data. And regardless of that, a document may still contain confidential company information.
For consultants, therefore, the better solution is often to separate the task from the client’s materials at first. For example, an AI can develop a general workshop structure without receiving the actual interview transcripts. Client-specific content is then added in the approved environment.
File contents should also be checked before uploading: comments, hidden worksheets, or additional information may reveal more than the visible first page.
Effective use of AI begins with clarifying the scope of the project. Consultants and client companies should work together to answer four questions:
A better-designed process might look like this: An interim manager makes use of the business environment that has been reviewed by the client. To draft a concept, he uses only the necessary, approved information. He then reviews the results from a technical perspective and supplements them using the original documents.
That takes more preparation than a spontaneous upload. But it gives you a chance to explain later what actually happened.
Any accidental upload should be investigated promptly. What content was uploaded? To which account? Which settings and connected services were active? What options are available for deletion or restriction?
The appropriate contacts should be notified in accordance with the agreed-upon procedure, and the measures taken should be documented. Simply deleting the visible chat does not prove that all traces of processing have immediately disappeared from the provider’s systems.
In the event of a personal data breach, a processor must notify the controller without delay. The controller shall assess, in particular, whether notification to the supervisory authority is required. For cases subject to notification, Article 33 of the GDPR requires immediate notification, preferably within 72 hours of becoming aware of the breach.
Not every instance of improper AI use automatically triggers a report to the authorities. However, an investigation of the specific incident is required.
Consultants do not need a specific legal ruling for every general question regarding wording. They need clear guidelines and a manageable approval process for sensitive cases.
Practical AI training programs can help convey these distinctions. An internal or external AI Officer can coordinate approved applications, responsibilities, and controls. This role serves as organizational support; it is not a blanket legal requirement for every company.
K11 Consulting integrates this governance work with data protection and information security. K11 Law Firm can provide additional legal review of contracts and confidentiality agreements.
AI can speed up the counseling process. However, it does not automatically expand the counselor's authority.
Anyone who wishes to use customer data should clarify in advance what information may be processed, in what environment, and for what purpose. Contract review, data protection, confidentiality, and technical oversight are all interrelated in this context.
The key question is not just, “Can AI turn this into a concept?” but also, “Am I allowed to give it this material?”