Yes, if the tool receives personal data about employees, customers or other people, GDPR Article 28 requires a written contract with the vendor before that data is processed. If staff only use the tool with no personal data, a data processing agreement is not needed, although a data-use policy still is useful.
Key takeaways
- A data processing agreement is needed before personal data goes into an AI tool that a vendor handles on your behalf.
- The vendor's role depends on what it does with the data, so read its terms and settings rather than assuming.
- A signed agreement should cover scope, instructions, subprocessors, data location, retention, deletion and breach notification.
- Personal accounts and consumer tiers can place team use outside any contract, even when the company has a business agreement elsewhere.
Do you need a data processing agreement when staff use an AI tool?
Usually yes, whenever the tool processes personal data for your company. The agreement should be signed before that data is entered, not after the rollout has started. The obligation comes from the General Data Protection Regulation (EU) 2016/679, where Article 28 sets out what a contract with a processor must contain.
This article is general information and not legal or tax advice. Your legal team should confirm how the rules apply to your sector and your data. Rui Codex provides technical implementation support, so the contract review itself belongs with qualified counsel.
Many teams miss this step because an AI tool feels like a text editor rather than a data service. Once staff paste customer emails, support tickets or employee notes into it, the tool is processing personal data. For a broader view of data protection duties in software projects, the security and GDPR articles on the Rui Codex blog cover related engineering decisions.
When does an AI tool count as a processor?
A vendor is a processor when it handles personal data on your behalf and under your instructions. The label depends on what the tool does with the data, not on how well known the product is. The same vendor can be a processor for one feature and a controller for another, so the analysis has to be done per use.
Questions that decide the role
- Does the vendor use your inputs only to deliver the service to your company?
- Does it use your data to train or improve models shared with other customers?
- Does it keep logs or outputs for its own purposes after the session ends?
- Does it engage other companies to handle the data, and are you informed?
If the vendor uses the data for its own purposes, it may act as a controller, and the arrangement needs a different legal assessment. The European Data Protection Board publishes guidance on these roles. The EU AI Act, Regulation (EU) 2024/1689, sets separate obligations for AI systems depending on their risk, but it does not replace the data protection duties that apply to personal data.
What should a data processing agreement cover for an AI tool?
It should describe the subject matter and duration of processing, the nature and purpose of the processing, the types of personal data, and the categories of people concerned. It should also list the processor's duties under Article 28(3), such as acting only on documented instructions and deleting or returning data at the end of the service.
- Instructions: what the tool may do with inputs and outputs, in plain terms.
- Subprocessors: a current list, notice of changes, and a right to object.
- Data location and transfers: where data is stored and what transfer safeguards apply.
- Retention and deletion: how long prompts, files and logs are kept, and how deletion is confirmed.
- Security measures: a description the vendor commits to, not a general promise.
- Breach notification: the timeline and the contact route for incidents.
- Assistance: help with data subject requests and impact assessments where needed.
Read the agreement against your actual use. A clause that covers the API may not cover the chat interface, and a business plan may carry different terms from a free plan. Teams that connect AI tools to business systems should also review how data moves between them, which is where system integration work usually changes the risk picture.
Hypothetical examples of team AI use
Here are two hypothetical scenarios:
Scenario one. A sales team pastes call notes with customer names and contract details into a summarising tool. The tool is used through a shared login under a consumer plan. No agreement covers the data, the retention period is unknown, and the team cannot say where the notes were stored. The fix is to move to a business account with a signed agreement and to stop using shared logins.
Scenario two. An operations team uses a chat assistant to reformat product descriptions that are already public on the website. No personal data is entered, so a data processing agreement is not required for this task. The team still writes a short rule that staff must not paste customer records or employee information into any tool that lacks a contract.
Both scenarios show that the question depends on the data, not on the brand of the tool. The same product can be fine for one task and unacceptable for another.
How do you check an AI tool before the first rollout?
Confirm the contract, the data flows and the account settings before any staff member receives access. Doing this in order avoids discovering a gap after personal data is already in the tool.
- List the tasks staff want to use the tool for, and mark which ones involve personal data.
- Identify the exact product tier and interface, since terms can differ between them.
- Obtain the vendor's agreement and check it against the clauses listed above.
- Confirm the answers on training use, subprocessors and data location in writing.
- Turn off features that store chat history or share data if the business does not need them.
- Issue company accounts only, and record who has access.
- Write a one-page usage rule with examples of what must never be entered.
- Review the setup after vendor terms or features change.
If your team connects the tool to a CRM, ERP or ticketing system, test that connection with dummy records first. For a structured review of the project itself, AI consulting support can help map the data flows before any build starts.
Where does this approach fail?
A signed agreement fails to protect you when the tool is used outside it. Personal accounts, browser extensions and copied prompts can move data to services no one has reviewed. A written policy without enforcement has the same weakness.
There is also a trade-off. Blocking all AI tools is simple but pushes staff toward unofficial use. Allowing everything is fast but leaves data flows unmapped. A middle path, with approved tools, clear rules and a short approval route for new tools, usually takes more upfront effort but holds up better in practice. Where the business needs a tailored automation layer rather than a general chat tool, AI automation design can keep personal data inside systems the company already governs.
Sources cited in this article
- General Data Protection Regulation (EU) 2016/679 (eur-lex.europa.eu)
- European Data Protection Board (edpb.europa.eu)
- EU AI Act, Regulation (EU) 2024/1689 (eur-lex.europa.eu)