Putting an AI on your support WhatsApp solves a lot, from group summaries to reply drafts. It also means your clients' conversations pass through one more system: names, phone numbers, addresses, an ID number sent "just for the invoice", a photo of a document, details of their business.
This post lays out what to consider about AI on WhatsApp and data protection before you switch anything on. It focuses on the EU's GDPR, which most other privacy laws resemble, and uses Brazil's LGPD as a second example. It is not legal advice; if you handle sensitive data or large volumes, talk to whoever owns privacy in your company.
Why data protection law applies here
The GDPR (Regulation (EU) 2016/679) covers almost any processing of personal data by a business that targets or operates in the EU. Reading, storing, transcribing and sending text to an AI model are all processing. A client's WhatsApp message nearly always contains personal data, even if it is only a name and a number. Brazil's LGPD follows the same logic, and so do many newer laws.
A few points weigh most in this case.
You are the controller. Whoever decides why and how the data is processed is the controller (GDPR Article 4(7)). If you use an AI tool to serve your clients, the main responsibility is yours, even if the processing runs on a third-party service acting as your processor (Article 4(8)). Processors need a contract that covers data protection (Article 28).
You need a lawful basis. Article 6 lists the bases. In client support, the usual ones are performance of a contract and legitimate interests. Consent is one option, not the only one, and often not the best fit.
Purpose limitation and data minimization. Article 5 asks you to use data for the purpose you collected it for, and only as much as you need. Using support chats to train a marketing model, for instance, is a different purpose.
Security. Article 32 requires appropriate measures to protect data against unauthorized access and leaks.
Automated decisions. Article 22 gives people rights when a decision with legal or similarly significant effects is based solely on automated processing. If the AI denies a request, applies a condition or classifies a client on its own, think about this one. Brazil's LGPD has a similar right to request a review (Article 20).
Data leaving the country. Many AI models run on servers abroad. International transfers have their own rules (GDPR Chapter V), and you need a valid transfer mechanism.
Transparency about AI. The EU AI Act adds duties for systems that interact with people, including making it clear when someone is dealing with an AI (Article 50). Even where no law requires it, being straight about it is the safer choice.
Where the messages actually go
The most important question when choosing a tool is simple: where does a message pass through, and where is it stored? There are usually three layers.
- Where conversations are stored. In the vendor's cloud, on your own server, or only on the phone.
- Where they go to be processed by AI. The model may run on your server or through an external API (OpenAI, Anthropic, Google and others).
- What the AI provider does with them. How long do they keep it? Do they train on it? Terms for paid APIs usually differ from free consumer chat apps.
Copying a client's conversation into a free consumer assistant is the path with the least control. Depending on the account settings, the content may be used to improve the service.
Questions to ask before choosing a tool
- Where are messages stored, and in which country?
- Which parts go to an external AI API, and which one?
- Does the AI provider use the data to train models? Can that be turned off?
- How long are messages and transcripts kept? Can they be deleted?
- Who on the team can see which conversations?
- Does the AI send messages on its own, or does someone approve first?
- If a client asks to access or delete their data (GDPR Articles 15 and 17), can you do it?
Practical safeguards for any tool
Human approval for decisions that matter. Refusing a request, charging a fee, changing contract terms. Let the AI draft and a person decide. This also helps with the automated-decision rules.
Less data in the chat. Ask for documents through a better channel when you can, and train the team not to ask for data they won't use.
Access per person. An agent sees the conversations they handle, not the whole number.
A record of who did what. Knowing whether a message came from the AI, from you or from a teammate helps you investigate problems and answer the client.
Transparency. If AI takes part in support, say so in your privacy notice. If a client asks whether they are talking to an AI, tell the truth.
How WACrew handles this
WACrew is open source and designed to run with you, on your machine or your own server. Conversations, audio files and transcripts live in a volume you control, not in a service run by the project. You can read the code and see exactly what it does with each message.
Some parts can use external services, and you choose which ones to turn on:
- Transcription runs locally on an Apple Silicon Mac, or through an API (OpenRouter) elsewhere.
- The copilot and AI drafting use the model you configure, through Anthropic, OpenAI or a compatible API.
- Some pre-send checks and video analysis use a model through an API.
On the control side, confirm-before-sending is the default, each team member has their own login with only the chats you allow, and every message records who wrote it.
There is also a risk outside privacy: WACrew connects through the WhatsApp Web protocol, not the official API. Our open-source WhatsApp API guide explains what that means.
FAQ
Can I use AI in WhatsApp support and stay GDPR compliant? Yes, with a lawful basis, a clear purpose, security measures and transparency. The GDPR doesn't ban AI. It asks you to take responsibility for the processing.
Do I need my clients' consent to use AI? Not always. Consent is one of the bases in Article 6, but contract and legitimate interests can also apply. Assess your own case.
Does a self-hosted tool solve compliance? It helps a lot with control over where data lives, but not with everything. Lawful basis, purpose, team access and what you send to external APIs are still your responsibility.



