Compliance and KVKK

AI and KVKK: What to Watch for in Prompts Containing Personal Data

Nanda Teknoloji Engineering Team · · 8 min read

In brief

Entering personal data into a prompt is a data processing activity under KVKK (Turkish Personal Data Protection Law), and every transmission to a provider must be assessed separately. As the data controller, you need to put purpose, legal basis, notice, data minimization, and security measures in place beforehand.

The moment you enter personal data into a prompt, Law No. 6698 on the Protection of Personal Data (KVKK) comes into play: sending information about an identified or identifiable person to an AI service is the processing of personal data, and transmitting it to the service provider requires a separate transfer assessment depending on where the provider is located. As the data controller, the organization must demonstrate the purpose, legal basis, notice obligation, and security measures for this processing. The practical result is this: every copy-paste done “just to be quick” falls within the organization’s accountability. This article is not legal advice; it gathers the principles and the controls that work in practice. For specific situations, consult your legal counsel and the publications of the KVKK Authority.

Which prompts contain personal data?

Most of the time it goes unnoticed. A name, an ID number, and a phone number are obvious signs, but the following content also carries personal data:

  • Customer complaint threads and support tickets
  • Résumés, performance notes, and disciplinary records
  • Signed contracts and party details
  • Error logs (IP address, user ID, email)
  • Transcripts extracted from meeting recordings
  • Any text containing special-category information such as health, union membership, or religion

Special-category data is subject to stricter conditions under the Law; prohibiting the sending of these data types to a general-purpose AI service by default is a sensible starting point.

How do the KVKK principles apply to the prompt scenario?

The general principles of the Law translate directly into control questions for prompt use.

Principle Question in the prompt context
Compliance with the law and rules of integrity Is this use within the data subject’s reasonable expectations?
Being accurate and up to date Could the information the model produces contain a wrong inference about a person, and who verifies it?
Processing for specified, explicit, and legitimate purposes Is the purpose for which the data was collected compatible with using AI?
Being relevant, limited, and proportionate to the purpose Is this much personal data really needed for the task?
Retention only for the necessary period How long does it remain with the provider and in the logs?

The “proportionality” principle is the most directly applicable one: you do not need the customer’s name and phone number to produce a summary. Data minimization here is both a legal and a technical gain.

The relationship between data controller, data processor, and provider

The organization sending the prompt is the data controller. If the AI service processes the data on the organization’s behalf and under its instructions, it acts as a data processor; if it uses the data for its own purposes (for example, model training), the allocation of roles and responsibilities changes. This distinction must be clear in the contract and terms of use. The main points to check:

  1. Does the provider use prompts and outputs for model training? Can this be turned off?
  2. In which countries is the data processed and stored?
  3. How are the retention period and deletion requests handled?
  4. Who are the sub-processors?
  5. What is the notification process in case of a data breach?

Transfer abroad is subject to separate provisions in the Law. Because transfer mechanisms and conditions may be updated, verify the current position from the legislative text and the Authority’s announcements. We discuss what choosing domestic, locally hosted infrastructure offers in this respect in our article on domestic cloud and data residency.

Notice and data subject rights

If personal data is processed with AI, check whether your notice texts cover this use. The employee notice is assessed separately for employee data, and the customer notice for customer data. You should also plan in advance how data subject requests (information requests, correction, deletion) will be reflected in AI records: can a person be searched for in prompt logs?

Decisions that produce outcomes about individuals solely through automated systems (for example, automated screening) require extra attention. For such uses, design human oversight and an avenue for objection from the start.

Security measures: What does the Law expect?

The Law obliges the data controller to take the necessary technical and administrative measures to ensure an appropriate level of security. In AI use, the practical equivalent is:

  • Access control: Authentication, role-based authorization, and a corporate account instead of an individual account.
  • Data minimization and masking: Removing personal data before it enters the prompt or replacing it with a placeholder; an AI DLP layer is used to apply this automatically.
  • Logging and traceability: Who sent which data to which tool? A central AI Gateway provides this record from a single point.
  • Policy and training: Explaining to employees, with concrete examples, what must not be sent; the corporate AI usage policy sets this framework.
  • Risk analysis: Running a short impact assessment for each new use scenario.

On the governance side, the NIST AI Risk Management Framework offers a structured method for identifying and monitoring risks.

Example scenario: The HR team’s résumé summaries

The HR team wants to use an external chat service to summarize incoming résumés. The résumés contain name, contact details, education, health status (sometimes), and a photo. The assessment proceeds as follows: is the use consistent with the candidates’ notice text, does the provider use the data for training, in which country is the data processed, and are all fields really needed? As a result, name, contact details, and photo are removed automatically, only the experience and skills text goes to the model, the output is verified by a person, and the screening decision stays with a human.

Common mistakes

  1. Assuming “the free version is fine.” The data use terms of individual and free accounts may differ from those of corporate contracts.
  2. Mistaking masking for anonymization. Data replaced with a placeholder may still be treated as personal data if it can be matched back.
  3. Not updating the notice text. If there is a new processing purpose and a transfer, the text should reflect it.
  4. Not separating special-category data. A separate, strict rule is needed for this content.
  5. Ignoring employees’ use of personal accounts. Uncontrolled use means the organization effectively fails to meet its security obligation.

Risk assessment by prompt use type

Not every use carries the same legal and technical weight. The organization can group use types into categories and set a default decision for each.

Use type Personal data status Default approach
General knowledge question, no personal data None Free; approved tool preferred
Text correction, drafting Occasionally incidental Employee training, short warning
Customer correspondence summary Present (name, contact, content) Masking mandatory, approved tool
HR, health, finance data Present, may be special-category Prohibited in general tools; only in an environment that has undergone legal assessment
Decision support about individuals Present, produces outcomes Human oversight, separate impact assessment

A classification like this gives employees a clear guide and, in the event of an incident, evidence that the organization took reasonable measures.

Topics to check in corporate contracts

When you decide to use an AI service at the organizational level, check the following topics in the contract and terms of use:

  1. Data use: Whether inputs and outputs are used for model development; the opt-out option.
  2. Retention and deletion: How long data is kept and how deletion requests are fulfilled.
  3. Location: The countries where processing and backup take place.
  4. Sub-processors: The third parties with whom the provider shares data, and notice of changes.
  5. Security: Encryption, access control, independent audit reports.
  6. Breach notification: How soon the organization will be informed in case of an incident.
  7. Audit right: The provider’s obligations when the organization or competent authorities request information.

If you cannot get a clear answer on all of these topics, limiting the tool to uses that contain no personal data is a sensible interim solution.

Data breach and incident management

An employee accidentally uploading a file containing personal data to an unapproved tool can trigger the incident management process. The Law provides for certain notification obligations in the event of a data security breach; verify the current legislation and the Authority’s announcements for deadlines and procedures. A short, pre-prepared incident procedure should include: who reports the incident, what information is collected in the first 24 hours (what was sent, to whom, and when), how the deletion request is communicated to the provider, and when legal counsel and the KVKK officer step in. A culture that punishes concealing an incident prevents this process from working.

Readiness checklist

  • Have the AI use scenarios involving personal data been inventoried?
  • Have the legal basis and purpose been documented for each scenario?
  • Do the notice texts cover this use?
  • Have provider contracts been reviewed against the topics above?
  • Is there a technical control that masks personal data before it enters the prompt?
  • Have employees been told with examples which data must not be sent?
  • Does the data breach procedure also cover AI incidents?

Decision summary

Assess AI use under KVKK with three questions: Which personal data is going out, and is it really necessary? Where and under what conditions does the provider process this data? How do I technically monitor this and prove it? If you do not have a clear answer to all three, start with data minimization and central oversight.

Moving forward with Nanda Teknoloji

Nanda Teknoloji carries out the discovery, design, migration, and operations steps in projects like this: mapping personal data flows, and setting up and operating the technical controls. The legal assessment is carried out together with the organization’s legal counsel. On the technical side, our AI security and AI DLP solutions support these controls.

Frequently asked questions

Does typing a customer's name into a prompt fall under KVKK?

Yes. Any information relating to an identified or identifiable person is personal data, and entering it into a prompt counts as data processing. Legal basis and compliance with the principles are therefore required.

Can anonymized data be entered into a prompt?

Data that has been truly anonymized is not personal data. However, masking that can be reversed or matched with other data does not by itself amount to anonymization; the method must be assessed carefully.

What happens if the provider is abroad?

The transfer of personal data abroad is regulated separately under KVKK. The provider's data processing location and conditions must be determined, and no transfer should take place without a legal assessment.

Does an employee's use through a personal account concern the organization?

If an employee sends organizational data to a tool through a personal account, the organization's data security obligation is engaged. That is why a usage policy and technical controls are needed together.

Related solutions

Related posts

Let's plan your critical systems together

Tell us what you need; our team will work out the right architecture and migration path with you.