Data protection and AI
AI in the workplace: when can mistakes justify dismissal?
Neither the use of AI nor an incorrect AI result is automatically a ground for dismissal. What is decisive is the specific duty, the permitted use, knowledge and fault. Clear rules, reasonable training and less severe measures are part of the assessment.

Telling four different conflicts apart
The employment law assessment depends on what is actually at issue: use of a system that has not been approved, disclosure of confidential data, unchecked adoption of incorrect results, or lack of the skills needed for a required application. These cases concern different duties and possible grounds for dismissal.
A general statement such as “ChatGPT is prohibited in the workplace” or “anyone may use AI” is of no help without knowing the contract, the company rules, the tasks and the functions used. An internal approval may also contain conditions on the scope of data and on oversight.
Who is responsible for the work result?
AI can support texts, calculations or research. Employees must nevertheless observe their specific duties of care and checking. How thorough a check is necessary and reasonable depends, among other things, on the task, the risk, the professional role, the instruction given and the foreseeable possibilities of error.
A fabricated quotation in an internal draft and unchecked use in a binding external document may be assessed differently. It is equally important whether an error was noticed, corrected in time or deliberately concealed. Not every mistake proves a serious culpable breach of duty.
Confidential information needs its own checkpoint
Uploading applicant data, health information, contract documents or trade secrets to external systems can affect data protection and confidentiality obligations. Technical access or a free user account is not legal approval. The permitted purpose, contractual conditions, scope of data and technical settings must fit the specific use.
Lack of AI competence is not the same as refusal to work
If a person cannot operate a required application safely, their skills and reasonable training must be considered. If they deliberately refuse a lawful, reasonable and sufficiently specific instruction, a different question arises. An employer must not hastily equate gaps in knowledge with an intentional breach of duty.
For a possible dismissal on personal grounds, the prognosis, the effect on the business and the alternatives must be assessed. Where the conduct is controllable, the role of a disciplinary warning as a less severe measure must as a rule also be clarified. The label “AI error” replaces none of these steps.
Fictional example from everyday office life
An employee uses an internally approved text tool for a first draft. Despite a required source check, the version sent contains incorrect information. To assess the case, the requirements, instruction, verifiability, fault and consequences, among other things, have to be established. Another case would be the deliberate upload of confidential client documents to an expressly prohibited external system. Without further facts, neither example leads automatically to a decision to dismiss.
What companies and employees should document
- Permitted applications, tasks and types of data.
- Specific steps for checking, labelling and approval.
- Training, queries and known limitations of the system.
- The error report, the correction and the actual consequences of an incident.
In the case of a disciplinary warning or dismissal, the system used and the exact use must be described. Sensitive inputs should not be reproduced again without protection.
Legal bases and sources
- § 106 GewO – Right of instruction
- Section 241 BGB – Duties of consideration
- § 1 KSchG – Reasons for termination
- § 626 BGB – summary dismissal
Legal position and source check: 7 November 2026. General information; assessing a specific case requires the complete documents.
For advice on your situation: Legal advice on AI and IT.
