Detailed analyses can be found in the publications under "Publications". For individual legal advice, please contact a law firm. The author assumes no liability for the accuracy and actuality of the information. For questions from students of the h_da, the author as data protection officer is responsible for data protection issues and the respective lecturers for questions regarding the use of data in examinations.

1.  Where does Copilot obtain its data from and what is legally problematic about it?

The data originates

• from ChatGPT, which uses data from the Internet and from licence data databases,
• from the Bing search engine,
• from Microsoft Graph, in which the data of the company using Copilot and the users in the company are collected, and therefore
• from the company itself.

See graphical overview at https://learn.microsoft.com/en-us/microsoft-sales-copilot/architecture 

Microsoft explains the possible uses at https://learn.microsoft.com/en-us/microsoft-365-copilot/microsoft-365-copilot-privacy:  

„Microsoft Copilot for Microsoft 365 provides value by connecting LLMs to your organizational data. Microsoft Copilot for Microsoft 365 accesses content and context through Microsoft Graph. It can generate responses anchored in your organizational data, such as user documents, emails, calendar, chats, meetings, and contacts. Microsoft Copilot for Microsoft 365 combines this content with the user’s working context, such as the meeting a user is in now, the email exchanges the user had on a topic, or the chat conversations the user had last week. Microsoft Copilot for Microsoft 365 uses this combination of content and context to help provide accurate, relevant, and contextual responses.“

It is not always clear from which source information originates and which input data is tracked and how (Microsoft explicitly points out that the policy settings that control the use of optional connected experiences in Microsoft 365 apps do not apply to Microsoft 365 Copilot and public web content).

The legal issues with this include

-> the quality assurance of results with multiple databases,

-> the combination of data from all Office programmes, which (naturally) does not comply with the data protection principles of data separation and data minimisation,
the planning of access authorisations and role models,

-> safeguarding the confidentiality of business secrets, even if Microsoft assures that company data will not be used for AI training via an "AI Safety Mechanism".

-> When introducing Copilot, the relevant legal steps must be checked, see details on the legal basis in "PROCESS OF INTRODUCING AI / USING COPILOT&CO IN THE COMPANY"

-> It must also be clarified to what extent instructions are to be delegated to the AI (independent decisions with legal effect are generally not permitted, Art. 22 GDPR).

-> The use is subject to co-determination, it is advisable to conclude corresponding works agreements and to clarify the obligation to use (§ 106 GewO) and authorisation under individual employment law.

2.  Who owns the copyright to the texts, images, presentations etc. generated by Copilot?

According to German law (§ 2 UrhG), only a HUMAN creation is protected by copyright, so that neither Microsoft as the provider of Copilot nor the user as the person who had the result generated by a so-called "prompt" is the author of the results. Nevertheless, Microsoft as the operator of Copilot as the party offering the service can determine to a certain extent the purposes for which the results are used. In its licence conditions, Microsoft transfers the rights to the results to the licensees.

-> When passing on data, you must check to what extent Copilot's own sensitive data is not inadvertently included.

3. May the texts, images and presentations generated by Copilot be passed on to third parties? Are there any risks involved?

Yes, as Microsoft freely licences the results of Copilot for all licensees. However, it cannot be ruled out that the rights of third parties whose work has been incorporated into the Copilot training database may be infringed, particularly in the case of image/video generation. In this case, however, you are generally entitled to indemnification from Microsoft.

In addition, care should be taken to ensure that the results do not include information on business secrets or sensitive (possibly personal) data that should remain confidential. Depending on the prompt and database, Copilot can access all sensitive data stored in your company's Office products.

-> When passing on data, it must be checked to what extent Copilot's own sensitive data is not inadvertently included.

-> If Copilot results are passed on externally without being checked, your own liability under the German Telemedia Act (TMG) or the principles of "Stoererhaftung" (Breach of Duty of Care) may come into consideration, especially if they are distributed via apps/websites.

4. May employees use the results generated by Copilot for professional purposes in order to save time?

Such use should always be disclosed to the employer. If you are paid for the creation of these results, concealing the use of Copilot could, in the worst case, constitute fraud against the person who paid for the personal creation of a text.

-> Usage guidelines should clearly define in which cases Copilot may be used and whether and how the results are to be labelled.

5. Is Copilot compliant with data protection law?

A distinction must be made between

a) the assessment of the general setup of Copilot as a Microsoft tool with potential data transfer to the USA and
b) the subsequent use for data management and data entry:

i)    As with any tool, the pure use is not data protection compliant if data about the user is not handled in a GDPR-compliant manner. The prompts generally provide a lot of information about the user (preferences, illnesses, political views, etc.), which may be stored together with other data (such as the user's IP address). Data protection issues may therefore arise even without the input of personal data. The Guidance of the Data Protection Conference ‘Artificial Intelligence and Data Protection’ Version 1.0 of 06.05.2024 , the checklist on the use of LLM-based chatbots issued by the Hamburg Data Protection Authority on 13 November 2023, and the Checklist for GDPR-compliant AI from the Bavarian Data Protection Authority of 24 January 2024 must be observed for individual data protection issues.

Necessary steps therefore include:

-> Data protection compliance of the tool

-> Data protection impact assessment

-> Involvement of the DPO

-> Scenario of use

-> Clarification of joint processing (if the use of the data by Microsoft is unclear, it is questionable whether order processing can exist)

-> Clarification of foreign data transfer

ii) If you include personal names in your prompts, you may expand Copilot's database to include personal data of third parties who cannot influence this. This is similar to Google search queries that combine people with certain characteristics and then lead to them being suggested as additions in the autocomplete (e.g. "Wilmer Ferrero corruption" or "Wilmer addicted to vegetables") because they have been stored in the Google database.

According to Microsoft, when a request is forwarded to Bing by employees, the company's data is forwarded, but not the user's account and client ID.
 

Usage guidelines must clearly define which data may be entered in Copilot and how personal data is to be handled, in particular with regard to

-> Legal bases

-> Data combinations

-> Sensitive data

-> Data disclosure

-> AI decisions / AI instructions

-> Role models
 

6. Information and materials

a) Data protection information

Depending on the purpose of use, the following must be taken into account during the initial review
- Checklist for the use of LLM-based chatbots from the Hamburg Data Protection Authority dated 13.11.2023

- The AI introduction model procedure before deployment

- The AI commandments in the subsequent handling of Copilot

b) Materials for practical use

aa) OpenAI

- Usage guidelines

- Enterprise Privacy Rules (including API and Model Training FAQ and the commitment to use customer data only with consent for training)

bb) Microsoft Copilot
- Microsoft Copilot data, privacy and security

- Managing access to public web content in Microsoft Copilot for Microsoft 365 

- Documentation for Microsoft Copilot for Microsoft 365

- Microsoft Purview privacy and compliance protection for Microsoft Copilot

- Microsoft 365 Security Conformance & Compliance Guide