Contents
At a glance
- Category
- Business and buying
- Legal basis
- Article 28 of the UK and EU GDPR
- Parties
- Controller (decides why) and processor (acts on instructions)
- In an AI stack
- Automation platform, model provider, messaging provider and logging tool are all processors
- Most important clause
- Sub-processors - it reveals where your data actually goes
- Keep
- A register of vendors, DPA versions, dates and signed copies
What a DPA is
A data processing agreement - also called a data processing addendum - is the contract between a controller, who decides why personal data is processed, and a processor, who processes it on the controller's instructions. Under UK and EU GDPR, Article 28 requires this contract to exist whenever a processor handles personal data on a controller's behalf. It is not optional and it is not a formality.
In an AI automation you are usually the controller for your customers' data, and your automation platform, model provider, messaging provider and logging tool are all processors. That means a DPA with each of them, and your own DPA with any client whose data you handle.
What Article 28 requires it to cover
The contract must set out the subject matter and duration of the processing, its nature and purpose, the type of personal data and categories of data subject, and the controller's rights and obligations. It must bind the processor to act only on documented instructions, to ensure confidentiality, to implement appropriate security measures, to assist the controller with data-subject rights and breach notification, and to delete or return the data at the end of the contract.
It must also govern sub-processors: the processor may not engage one without authorisation, and must impose equivalent obligations on them. That clause is the one that matters most in AI, because your automation platform's sub-processor list is where you discover which model providers and infrastructure vendors your customer data actually reaches.
Getting one in practice
Most serious vendors publish a standard DPA you accept or countersign, often in the account settings or the legal section of their site. The work is not negotiating it - for a small business that is rarely realistic - but reading it for three things: the sub-processor list and how changes are notified, the transfer mechanism for anything leaving your region, and the security commitments.
Then keep a register: which vendors process personal data for you, which DPA version you accepted, when, and where the signed copy lives. That register is what a client's procurement team asks for, and assembling it retrospectively during a security review is considerably less pleasant than maintaining it.
Where AI complicates it
Three specific issues. Training on your data: a processor using your data to improve its own models is processing for its own purposes, which is not what a controller instructed. Check the terms and prefer providers that contractually exclude API data from training - see zero data retention.
Retention: providers commonly retain inputs for a period for abuse monitoring, which is a legitimate purpose but must be disclosed and bounded. Sub-processor churn: AI vendors change infrastructure providers more often than traditional software vendors, so the notification mechanism for sub-processor changes is worth reading rather than skimming.
What you should know
This is orientation, not legal advice - get advice appropriate to your jurisdiction and the data you handle. What you can do without a lawyer is the groundwork: know your data flows, know which vendors touch personal data, hold the DPAs, and keep the register current. That preparation makes any legal review cheaper and faster.
It also intersects with data residency and with your logging practice, since traces containing customer text are personal data processing like any other and belong in the same register.
DPA FAQ
What is a DPA?
A data processing agreement - the contract required between a controller, who decides why personal data is processed, and a processor acting on their instructions. Article 28 of the UK and EU GDPR makes it mandatory whenever a processor handles personal data on your behalf.
Do I need a DPA with my AI provider?
If customer personal data reaches them, yes. In a typical AI automation your model provider, automation platform, messaging provider and logging tool are all processors, so you need a DPA with each - and your own with any client whose data you handle.
What should a DPA cover?
The subject matter, duration, nature and purpose of processing, the data types and data subjects, processing only on documented instructions, confidentiality, security measures, assistance with data-subject rights and breaches, deletion at the end, and sub-processor obligations.
Why do sub-processors matter so much?
Because that list is where you discover which model providers and infrastructure vendors your customer data actually reaches. AI vendors change infrastructure more often than traditional software vendors, so how changes are notified is worth reading rather than skimming.
How do I get a DPA from a vendor?
Most publish a standard one you accept or countersign in account settings or the legal section of their site. Negotiating it is rarely realistic for a small business, so the work is reading it for the sub-processor list, the transfer mechanism, and the security commitments.
Does my logging tool need a DPA?
Yes, if traces contain customer text - and they usually do, since the whole point of AI observability is capturing the prompts as sent. It is one of the most commonly forgotten entries in a processor register precisely because it feels like infrastructure rather than data.
Primary source
Checked August 2026. Anything priced or rate-limited moves, so verify against the source before you budget.
Want this built rather than explained?
Knowing what dpa means is one thing; having it running in your business is another. We scope one workflow at a time, quote a fixed price in writing after a free consult, build it, and hand it over with documentation you own.
