Mexico's Federal Law on the Protection of Personal Data Held by Private Parties (LFPDPPP), published on March 20, 2025, does not mention artificial intelligence, but it applies to any processing of data, manual or automated. If you use AI with your customers' data, your company is still the responsible party: you need a privacy notice that covers that use, security measures, a provider that does not train on your data, and a person who reviews automated decisions that affect your customers. Fines go up to 320,000 times the UMA.
What does Mexican law say about using AI with customer data?
Nothing specific, and that does not excuse you. The current law defines "processing" as any operation on personal data carried out by manual or automated means: obtaining, using, storing, transferring. Sending a customer's history to an AI model so it drafts a reply is processing.
The law requires you to respect eight principles (lawfulness, purpose, loyalty, consent, quality, proportionality, information and accountability) and names as the authority the Secretaría Anticorrupción y Buen Gobierno (Ministry of Anti-Corruption and Good Governance), which took over the functions of the former INAI in this area. Its latest amendment was published on November 14, 2025.
On the regulations: as of November 2026, the Chamber of Deputies' list of regulations in force still shows the Regulations of this law published in 2011, and we did not find a new one. Check with your legal advisor which parts apply.
This article expands on step 2 of how to implement AI without taking risks. It is not legal advice: use it to arrive prepared when you meet with your lawyer.
What personal data should you not give to an AI?
Sensitive data, unless you have a legitimate reason and the right consent. The law calls data sensitive when it touches the most intimate sphere of a person or when its misuse can lead to discrimination or serious risk. To process it you need the data subject's express written consent, with a handwritten signature, an electronic signature or an authentication mechanism.
In practice, for a distributor or a manufacturer: employee health data, payroll files with medical information, or any data that reveals ethnic origin, beliefs, political opinions or sexual orientation should not go through an AI assistant without a prior analysis. And in general, apply proportionality: if the AI only needs the order number, do not send it the customer's full file.
What it costs you: fines for violations involving sensitive data can be doubled.
Is the AI provider a "processor" or a "third party" under the law?
It depends on who the provider is working with the data for. The law distinguishes two roles:
- Processor (persona encargada): processes the data on behalf of your company. This is the case of an AI provider that processes your conversations only to deliver the service to you.
- Transfer: disclosing data to someone other than the data subject, your company or the processor. If the provider uses the data for its own purposes (for example, to improve its models), it looks more like this. To transfer data, the law requires you to communicate the privacy notice and the purposes, and the recipient takes on the same obligations you have.
That is why what the provider's contract says about training and retention matters so much. It is not a technical detail: it defines which legal role you are in.
What should your privacy notice say if you use AI?
The same as always, but updated to the actual use. The law requires the notice to include, at a minimum: who you are and your address; what data you process, identifying the sensitive data; the purposes, distinguishing those that require consent; how the data subject can limit use; how to exercise the rights of access, rectification, cancellation and objection (ARCO); and how you will notify them of changes.
If you are going to use AI for a new purpose (for example, analyzing WhatsApp conversations to qualify prospects), check that it is covered. If an agent serves your customers on that channel, also review the WhatsApp rules for AI agents.
Can your customers object to decisions made by an AI?
Yes. It is probably the most important point for anyone using AI in credit, collections or hiring. The law gives the right to object when a person's data is subject to automated processing that produces unwanted legal effects or significantly affects their interests, and that is intended to evaluate them without human intervention: their financial situation, reliability, behavior, among others.
Translated: if an AI alone decides who gets credit or whose collections get escalated, that customer can object. The practical way out is for a person to review and sign off on those decisions. If you use Claude, Anthropic's Usage Policy taking effect on November 12, 2026 goes in the same direction: for high-risk uses such as finance or employment, it requires a person with the authority to change the recommendation and that the affected person know AI was used.
What should you check with the AI provider before sending it data?
Four questions: does it train on your data, how long does it keep it, who else sees it, and can you delete it? The answers change by product, even within the same provider:
| Product | Does it train on your data? | Published retention |
|---|---|---|
| Claude for business and API (Anthropic) | Not by default, unless you submit feedback or authorize it | API: deleted within 30 days; conversations flagged for safety, up to 2 years |
| Claude Free, Pro or Max (personal accounts) | The user chooses | If allowed, up to 5 years, de-identified |
| WhatsApp Business Platform (Meta) | The terms prohibit using that data to train AI, except for a model exclusive to you | Check Meta's terms |
| Meta Business Agent | Meta may use the content to improve its AI models | Check its terms |
| Odoo 20 AI agents | Check with Odoo and with the model provider (OpenAI or Gemini) | Conversations with agents: up to 30 days after the last message |
Data checked in November 2026; providers change their policies. The practical conclusion: have your team use business accounts, not personal ones, when working with customer information. More detail on Claude in what's new from Anthropic in 2026 and on Odoo in the AI that already comes built into Odoo 20.
What fines apply for mishandling personal data?
From 100 to 320,000 times the Unidad de Medida y Actualización (UMA, Mexico's unit of measure and update), depending on the violation, with an additional fine for repeat offenses and double for sensitive data. The law also establishes crimes: three months to three years in prison for anyone who, authorized to process data and for profit, causes a security breach.
In addition, the law requires you to promptly inform the data subject of security breaches that significantly affect their property or moral rights. If an AI provider suffers a leak involving your customers' data, as the responsible party, the notification is up to you.
What concrete measures should your company take?
The law requires administrative, technical and physical security measures, proportional to the risk and the sensitivity of the data. The technical reference frameworks, such as NIST's generative AI profile and the OWASP Top 10, agree on the basics. In practice:
- Inventory. Write down which AI tools your team uses, with what data and with which account.
- Business accounts. Prohibit the use of personal AI accounts with customer data.
- Minimum data. The AI sees only what it needs for the task. OWASP recommends limiting access to sensitive data with the principle of least privilege.
- Contract. Check that the provider does not train on your data and how long it keeps it.
- Privacy notice. Update it to cover the actual use of AI.
- Person in the loop. Decisions that affect credit, collections or employment are reviewed by a person.
- Confidentiality and audit trail. The law requires anyone who processes data to keep it confidential; have every AI action logged in your system.
If your data lives in scattered spreadsheets, putting it in a single system with per-user permissions makes almost all of the above easier; it is part of an Odoo implementation.
Glossary
- Responsible party (responsable)
- The company that decides on the processing of personal data. If you use AI with your customers' data, that is you.
- Processor (persona encargada)
- Whoever processes data on behalf of the responsible party, such as a service provider.
- Sensitive data
- Data that touches the most intimate sphere of a person or whose misuse can lead to discrimination.
- ARCO rights
- Access, rectification, cancellation and objection: what the data subject can request about their data.
- UMA
- Unidad de Medida y Actualización, a peso-denominated reference used to calculate fines.
Frequently asked questions
Does the personal data law prohibit using artificial intelligence?
No. The law does not mention AI, but it applies to any processing of data, manual or automated. You can use it if you comply with its principles and obligations.
Do I need my customers' consent to use AI with their data?
It depends on the purpose and the type of data. For sensitive data, the law requires express written consent. For everything else, check with your advisor whether your privacy notice covers that use.
Who is the personal data authority now that the INAI no longer exists?
The Secretaría Anticorrupción y Buen Gobierno, under the law published on March 20, 2025.
Can I use ChatGPT or Claude for free with customer data?
It is not advisable. On personal accounts the training and retention terms differ from business ones, and your company remains responsible for that data.
What do I do if my AI provider suffers a data breach?
The law requires you to promptly inform the data subjects who are significantly affected. Have a written procedure before it happens.
Sources: Ley Federal de Protección de Datos Personales en Posesión de los Particulares (DOF 20-03-2025, last amended 14-11-2025), arts. 2, 5, 8, 15, 18, 19, 20, 26, 35, 59 and 62 · Cámara de Diputados, historial de reformas de la LFPDPPP · Cámara de Diputados, reglamentos de leyes federales vigentes, accessed November 2026 · Anthropic Privacy Center, "Is my data used for model training?", August 2026 · Anthropic Privacy Center, "How long do you store my data?", July 2026 · Anthropic Privacy Center, "How long do you store my organization's data?", July 2026 · Anthropic, "2026 Usage Policy update", October 2026 · Meta, "Meta Terms for WhatsApp Business Platform", section 4.7, September 2026 · Meta, "Meta Business Agents and Platform Terms of Service", July 2026 · Odoo, "Odoo 20 Release Notes", September 2026 · OWASP, Top 10 for LLM Applications 2025: sensitive information disclosure · NIST AI 600-1, generative AI profile, July 2024
