Can I Safely Give AI My Customers' Data? A Privacy Checklist
Before giving an AI tool customer data: what Saudi PDPL says, transfers outside the Kingdom, what never to paste, and a checklist to vet the AI provider.
Yes, with conditions: you need a lawful basis for processing, such as consent or fulfilling the order; a business AI tool that does not train its models on your data, under a clear processing agreement; compliance with the rules on transferring data outside Saudi Arabia; and minimal data with sensitive details masked. Never paste customer data into a free public chatbot.
01The short answer: not all AI tools are equal
| Tool type | Training on your data | Suitable for customer data? |
|---|---|---|
| Free or personal public chatbot (such as a personal ChatGPT account) | According to OpenAI, conversations are used for improvement by default unless you turn it off in data controls | No |
| Business edition (such as ChatGPT Business, Enterprise or the API) | According to OpenAI, no training on business data by default | Possibly, after vetting the provider and the contract |
| Messaging platform with a built-in AI agent | Depends on the provider's contract and policies | Yes, if it passes the checklist below |
Source: OpenAI's enterprise privacy page and its Help Center article on how data is used to improve models, September 2026.
02What does Saudi Arabia's Personal Data Protection Law say?
- A lawful basis: process data with the customer's consent, to perform a contract with them, or for a legitimate interest after assessing it, and for a specific purpose.
- Data minimisation: collect and send to the tool only what the purpose requires.
- Sensitive data: such as health, genetic, biometric and credit data, beliefs and ethnic origin. It needs stronger protection, and often explicit consent and an impact assessment.
- A contract with the processor: a provider that processes data on your behalf needs a contract that obliges it to protect the data, notify you of any breach, and not engage a sub-processor without your approval.
- Breaches: notify SDAIA within 72 hours of becoming aware of the incident, and notify data subjects without undue delay when needed.
- Penalties: fines can reach 5 million SAR, and the law has been enforced since 14 September 2024.
This is a general summary, not legal advice; check the official text or a legal adviser.
03Transfers outside the Kingdom: where is your data processed?
Most AI models run on servers outside Saudi Arabia, which counts as "transferring personal data outside the Kingdom". The data transfer regulation, amended in 2024, allows transfers under conditions, including:
- A country or entity with an adequate level of protection, as assessed by SDAIA.
- Appropriate safeguards: standard contractual clauses, binding corporate rules for groups, or a certification.
- A risk assessment when transferring under appropriate safeguards, or when transferring sensitive data on a continuous or large scale.
Ask the provider two direct questions: where is the data stored and where is it processed? And will they sign standard contractual clauses or an equivalent? A vague answer is a red flag.
04What should you never paste into a public chatbot?
- National ID, residency (iqama) and passport numbers.
- Card and bank account numbers.
- Medical reports, prescriptions or any health information.
- Payroll records or credit data.
- A full customer list with names and numbers so it can write "a message for each one".
- Entire WhatsApp conversations without masking names and numbers.
- Contracts and special prices for key accounts.
SDAIA's generative AI guidelines urge organisations to take a firm stance against employees entering confidential information into external generative AI tools. Write a one-page internal policy and train your team on it.
This is an example of a safe prompt: it gives the tool the context without the customer's name, number or order number.
05Checklist before choosing an AI provider
| Question | Why it matters | Reassuring answer |
|---|---|---|
| Do you train your models on our data? | Your data could appear in other people's outputs | No, written into the contract |
| Where is the data stored and processed? | Rules on transfers outside the Kingdom | A named country and signed safeguards |
| How long is data kept? Can I delete it? | Data minimisation and the customer's rights | A defined period and deletion on request |
| Who on your side can access the conversations? | Internal access | Limited permissions and an access log |
| Will you sign a data processing agreement? | A legal obligation on the processor | Yes, with a list of sub-processors |
| How will you notify us of a breach? | 72-hour deadline to notify SDAIA | Immediate notice within a defined time |
| Is the data encrypted? | Protection in transit and at rest | Yes, in transit and at rest |
| Do you comply with Meta's terms for WhatsApp data? | Meta bans using Business Platform data to train general models | Yes, and no training except for a model exclusive to you, if any |
Source: WhatsApp Business Solution Terms, last updated March 2026: Business Platform data may not be used to train or improve AI models, except to fine-tune a model for your exclusive use.
06Masking and minimising: practical steps
- Replace names and numbers with placeholders before any external analysis: [Customer 1], [Mobile].
- Send the agent only the fields it needs: an order number is usually enough instead of a full address.
- Set the agent never to ask for sensitive data, and to hand over to a staff member if the customer sends it.
- Delete sensitive attachments from conversations once their purpose is served, according to your retention policy.
- Review a sample of conversations every month to make sure no data is leaking in replies.
The best-protected data is data you never collected. Before adding any new field to a form or chat, ask: do I really need it?
07Inside your team, and with Modeer 360
Half of all privacy risk is internal: customer chats on employees' phones, and an employee who leaves the company taking customer numbers with them. In Modeer 360, conversations live in a unified inbox on the company's official number, each employee signs in with their own account, and the customer record stays in the CRM.
For details on where data is stored, how long it is kept, and the data processing agreement: ask the Modeer 360 team and get it in writing, as you would with any provider.
Key takeaways
- Never paste customer data into a free public chatbot.
- Choose a business tool that does not train on your data, with a clear processing agreement.
- Know where your data is processed and follow the rules on transfers outside the Kingdom.
- Minimise data and mask sensitive details before any AI processing.
Frequently asked questions
Is using ChatGPT to write customer replies against the law?
Writing a generic reply with no personal data is fine. The problem is pasting customers' names, numbers and details into a personal tool that trains on conversations by default.
Can my customers' data be processed outside Saudi Arabia?
Yes, under the conditions of the regulation on transfers outside the Kingdom, such as an adequate level of protection or appropriate safeguards like standard contractual clauses, with a risk assessment in specific cases.
What should I do if data leaks through an AI tool?
Cut off access, assess the impact, notify SDAIA within 72 hours of becoming aware, notify affected customers when needed, and consult a legal specialist.
Where is conversation data stored in Modeer 360?
Ask the Modeer 360 team about the storage location, retention period and data processing agreement, and request the answer in writing.
- #Personal Data Protection Law (PDPL)
- #Data transfers outside Saudi Arabia
- #Sensitive data
- #Vetting an AI provider