Your team pastes customer lists into AI chats from personal accounts. What Law 1581 and the SIC say, what AI vendors do with that data and what to check this week.
Can you put customer data into ChatGPT or another AI?
It is Tuesday and your collections coordinator has dozens of payment reminders to send. She exports the list of overdue customers from the system, with name, ID number (cédula), phone and balance, pastes it into ChatGPT from her personal account and asks for a message for each one. At the next desk, someone in sales uploads the prospect spreadsheet to Gemini to segment it. And last week a vendor offered you a WhatsApp agent that will read every chat.
Nobody means any harm. The problem is that nobody knows where that data went, whether the provider uses it to train its models, or whether your customers ever authorized that use. If one of them asks tomorrow what you did with their ID number, your company has to answer, not the tool.
Our WhatsApp and invoice guides touched on this briefly. This is the full version.
Why does this happen in so many small businesses?
- Everyone uses a personal account, free or paid with their own card, with no contract with the company.
- The data policy is an old PDF written to tick a box, and nobody connects it to AI.
- The consent customers signed does not cover these uses. It says "billing" or "commercial purposes", not "technology providers abroad".
- Nobody knows what sensitive data is, so a medical record or a photo of an ID card ends up in a chat.
The law is not new. Law 1581 of 2012 and its Decree 1377 of 2013, now compiled into Decree 1074 of 2015, apply to any tool. What is new is how easy it is to paste an entire database into a text box.
What counts as personal data, and what is sensitive data?
Personal data is any information linked to an identified or identifiable individual: name, ID number, phone, email, address, purchase history, outstanding balance.
Sensitive data (Article 5 of the law) is data that affects someone's privacy or whose misuse could lead to discrimination: health, sex life, biometric data such as fingerprints, face or iris, racial or ethnic origin, political views, religious beliefs, and membership in unions or social organizations. As a rule, processing it is prohibited except in specific cases such as the individual's explicit consent, which they are not required to give.
Of the principles in Article 4, these matter most when you use AI:
- Purpose: you use data for what you told the customer. If you collected it for billing, analyzing it with AI for something else is a new purpose.
- Freedom: processing requires prior, express and informed consent, except where the law says otherwise.
- Restricted access and circulation: only authorized people get to see the data.
- Security and confidentiality: technical, human and administrative measures to prevent loss or unauthorized access or use.
Customers also have rights (Article 8): to know, update and correct their data, request proof of consent, be told how their data is used, revoke consent or ask for deletion, and file complaints with the SIC. If you do not know which AI tools hold their data, you cannot answer them.
What does SIC Circular 002 of 2024 require for AI?
The Superintendence of Industry and Commerce (SIC), Colombia's data protection authority, issued it on August 21, 2024. It is not a new law: it explains how the data protection regime applies to anyone who develops or uses AI systems with personal data. That includes a small business that only uses a chat tool or buys an agent. The core points:
- Four criteria. Processing must be suitable (it serves the goal), necessary (no less intrusive measure works as well), reasonable (it pursues a legitimate aim) and proportionate (the benefit justifies the risk to the individual).
- Precaution. If there is no certainty about the harm processing could cause, the company must refrain or take preventive measures.
- Privacy impact assessment. When a high risk to individuals is likely, it must be carried out and documented before the system is designed.
- Quality. Data must be truthful, complete, accurate, up to date, verifiable and understandable.
- Privacy by design and accountability. Complying is not enough: you must be able to prove it with documents.
At the office, that means asking: does the AI really need the customer's ID number to draft a payment reminder? Almost never.
Is sending data to an AI the same as handing it to a third party?
Yes. The data leaves your company for a provider that almost always processes it outside Colombia. Colombian rules distinguish two cases:
- Transmission (transmisión): you send data to a processor that handles it on your behalf and under your instructions.
- Transfer (transferencia): you send data to another controller, which uses it for its own purposes.
With a business plan or the API, the provider processes on your behalf under a contract, which looks like a transmission. Decree 1377 allows international transmissions without notifying the individual or asking for consent when there is a contract that sets the scope of processing, the processor's activities and its security and confidentiality duties. The provider's data processing addendum (DPA) may cover part of that; check with your lawyer whether it meets Colombian requirements.
With a personal account there is no contract with your company, and the provider may use the content for its own purposes, such as training models. It is hard to argue it is acting only on your behalf.
As for the destination country: Article 26 of the law prohibits transfers to countries without an adequate level of protection, with some exceptions. In 2017 the SIC published a list of countries with an adequate level (External Circular 005), which includes the United States and the European Union countries. In December 2025, through External Circular 003, it adopted model contractual clauses for international transfers and transmissions, which are optional.
What do OpenAI, Anthropic and Google do with your data?
As of October 2026, according to each provider's documentation:
- OpenAI (ChatGPT): by default it does not train on data from ChatGPT Business, Enterprise or the API. On personal Free, Plus and Pro accounts, training is on by default; you turn it off under Data Controls, "Improve the model for everyone".
- Anthropic (Claude): it does not train on data from its commercial products (team and enterprise plans, and the API), and its DPA is built into the commercial terms. On personal Free, Pro and Max accounts the user decides; if they allow training, chats may be kept for up to five years.
- Google (Gemini): with Google Workspace, content is not human-reviewed or used to train models outside your organization without permission. In the Gemini app with a personal account, human reviewers may read some conversations, and Google advises against entering confidential information.
These terms change. Read the provider's current page and save a dated copy: that is also accountability.
What should you check this week before using AI with customer data?
- Day 1: take inventory. Ask, without blame, which AI tools each team uses, with which account and with what data.
- Day 2: sort data into three levels. What can go into any approved tool (text with no personal data), what only goes into tools under contract (basic personal data), and what goes into no AI without legal review: sensitive data, medical records, biometrics, children's data, and ID numbers combined with financial data.
- Day 3: review your data processing policy and privacy notice. Do they mention the purposes you would use AI for, and that you share data with technology providers, including outside Colombia? If not, update them; a new purpose may require asking for consent again.
- Day 4: review your providers. Business plan or API, DPA accepted, whether they train on your data, how long they keep it and where they process it. Put it all in one folder.
- Day 5: write a one-page rule. Approved tools, which data goes to which one, and who handles customer requests and complaints.
What about the National Database Registry (RNBD)? According to the SIC, only companies and non-profits with total assets above 100,000 UVT, plus public entities, must register. Micro and small businesses do not, but every other obligation still applies: policy, consent, a complaints channel and security.
What are the risks and limits?
- Penalties are real. Article 23 of Law 1581 lets the SIC impose fines of up to 2,000 monthly minimum wages, suspend processing for up to six months and order closures. In October 2025 it ordered World (formerly Worldcoin) to shut down its Colombian operation for collecting iris data without meeting the law, and in June 2026 it upheld that decision, including an order to delete the data.
- A good contract does not fix everything else. A business plan does not replace your customers' consent or an up-to-date policy.
- Anonymizing is not always enough. A job title, a neighborhood and a purchase date can identify someone even without a name.
- Provider terms change. Review them every time you renew.
- Decisions about people stay with people. AI can draft or summarize; deciding who gets credit or who gets hired should not be left to it alone.
This is general information, not legal advice. If you handle sensitive data or large databases, review your case with a data protection lawyer before connecting any tool.
Where should you start?
The goal is not to ban AI. It is to give every piece of customer data a clear path: which tool touches it, under which contract and for what. At 77Rentals we built that in from the design stage: contracts signed online carry the data protection (habeas data) notice inside the same signing flow.
With this week's inventory and data levels, you know where you stand. The rest is sorting out contracts, wording and habits.