How to use ChatGPT or Claude with your company data without losing control

In many small businesses, ChatGPT or Claude is already part of the working day, even if everyone uses it their own way. Someone pastes a client’s email to write a better reply; someone else uploads a spreadsheet to get a quick summary. It works, but the business has no idea what data is leaving, where it goes or who is accountable for it.

Getting this in order does not take a legal department. It takes deciding what the assistant can see, what it can do and who oversees it, and writing that down. This guide is written for businesses operating in Spain, so it covers EU rules plus Spanish guidance from the AEPD, the Spanish Data Protection Agency. It is up to date as of October 18, 2026, cites the official texts and is general information, not legal advice.

What connecting an AI assistant to your company data means

There are two ways to give an assistant your data. The first is the one almost everyone uses: copy and paste, or upload a file into the chat. The second is to connect it: with your authorization, you give it access to a business tool, such as your CRM, project tracker or calendar, so it can look things up or act there when you ask.

Those links are called connectors. Many of them use MCP (Model Context Protocol), an open standard Anthropic introduced in late 2024 that Claude and ChatGPT, among others, now support. Think of MCP as a common plug: the tool announces what it can do (find a client, create a task, read an invoice) and the assistant uses those actions when it needs them to answer you.

The difference matters. When you paste something, you choose what leaves and when. When you connect a tool, the assistant can request data on its own while it works on your request and, if the connector allows it, also write, create or change records. So treat a connection the way you would treat handing over a key: to whom, for which doors and for how long.

Seven things to check before you connect anything

1. What it sees and what it can do

Look at the access the connector asks for: read-only or read and write, which modules, which data. Start with the minimum and widen it when you need to. GDPR calls this data minimization: processing only data that is adequate, relevant and limited to what each purpose needs (Article 5(1)(c)).

2. Which account it uses and how you cut it off

It is better for each person to connect with their own account, through a standard authorization method like OAuth, than to share passwords or access keys. Check that access can be withdrawn at any time, and who can do it.

3. What gets logged

Ask whether there is a record of what the assistant looked up or changed, and at whose request. The AEPD’s February 2026 guidance on agentic AI (in Spanish) puts the emphasis on traceability: being able to tell what was done, when, who asked for it and with which data.

4. What contract you have with the provider

If the assistant will process personal data about clients, employees or suppliers, the AI provider acts as a data processor, and the relationship has to be governed by a contract (Article 28 GDPR). Check whether your plan includes one, because not every plan does. The AEPD publishes guidelines for controller-processor contracts (in Spanish) that help you review it.

5. What the provider does with your conversations

Check the terms and the settings: can the provider use what you type to train or improve its models, can you switch that off, and how long are conversations kept? The AEPD advises reading the terms of an AI service before using it and choosing the most privacy-protective versions.

6. Where the data is processed

GDPR does not require data to stay in the European Union, but if it leaves the European Economic Area, the transfer needs a basis: an adequacy decision from the European Commission or appropriate safeguards such as standard contractual clauses (Articles 45 and 46). Ask the provider where it processes data and under which safeguards.

7. What never goes into a chat

Health data, bank details, ID documents, data about minors or a client’s confidential information do not go into a chat without a clear reason and a legal basis. The AEPD sums it up in its January 2026 ten-point guide, “Cuidado con lo que le confIAs” (“Careful what you confide in AI”): at work, follow your organization’s policies and do not include confidential data about the organization, its staff or its clients. If you only need the reasoning, describe a made-up case.

What GDPR requires today

The General Data Protection Regulation has no separate chapter for AI: it applies as it does to any other processing. When you use an assistant with personal data, these are your questions:

• Legal basis (Article 6): on what basis are you processing that data for that purpose?
• Minimization (Article 5(1)(c)): does the assistant need all of it?
• Processor (Article 28): is there a contract with the provider?
• Security (Article 32): who has access, through which account, and what is logged?
• Impact assessment (Article 35): if the processing is likely to result in a high risk to people, assess it beforehand. The AEPD notes that using AI agents does not, on its own, make an assessment mandatory: it depends on the processing.
• Automated decisions (Article 22): people have the right, with exceptions, not to be subject to a decision based solely on automated processing that significantly affects them. Rule of thumb: the AI suggests, a person decides.

If your company has elected worker representatives, Spain adds one more rule. Article 64.4.d of the Workers’ Statute entitles them to be informed of the parameters, rules and instructions behind algorithms or AI systems that affect decisions on working conditions and on access to and retention of employment, including profiling.

What the EU AI Act already requires from businesses that use AI

The EU AI Act, Regulation (EU) 2024/1689, applies in stages, and in July 2026 it was amended by Regulation (EU) 2026/1744, known as the Digital Omnibus on AI. If your business uses ChatGPT or Claude for work, the Act calls you a “deployer”: anyone using an AI system under its own authority, outside purely personal, non-professional use (Article 3). Here is what applies to you as of October 18, 2026.

Since February 2, 2025: AI literacy and prohibited practices

AI literacy (Article 4). Since July 27, 2026, as amended, the article says providers and deployers shall take measures to support the development of AI literacy among their staff and others using these systems on their behalf, taking into account their knowledge, experience and the context of use. It does not require you to guarantee any particular level for each person, but it does require measures: short training, usage rules, examples of what to do and what not to do.

Prohibited practices (Article 5). The one closest to an ordinary business: using AI systems to infer the emotions of a person in the workplace is banned, except for medical or safety reasons (Article 5(1)(f)). Do not use tools that read your team’s mood from their voice or face.

Since August 2, 2026: transparency

If you use AI to generate or manipulate images, audio or video that could pass as real (a deep fake, in the Act’s terms), you have to disclose that it is artificial. The same applies if you publish AI-generated text to inform the public on matters of public interest, unless it has gone through human review or editorial control and someone holds editorial responsibility (Article 50(4)).

From December 2, 2027: high-risk uses

Some uses count as high-risk and bring extra obligations for deployers: assigning human oversight to people with the necessary training and authority, keeping the logs the system generates, and informing workers and their representatives before using it in the workplace (Article 26). For a small business, the most likely ones are in employment: using AI to recruit (targeted job ads, filtering applications, evaluating candidates) or to make decisions on working conditions, promotion or termination, to allocate tasks based on behavior, or to monitor and evaluate how people work (Annex III, point 4).

Those obligations were due to apply on August 2, 2026; the Digital Omnibus pushed them back to December 2, 2027, for Annex III uses. If you already use AI in any of these processes, do not wait: have a person review and decide, and document how you do it. That helps you today under GDPR and Spanish employment law, and later under the AI Act.

A plan for the first week

You do not need a months-long project. One well-used week takes you from everyone doing their own thing to a clear set of rules:

1. Monday: take stock. Ask your team which assistants they use, with which account (personal or company) and for what. No blame: the point is to know.
2. Tuesday: pick the tool and plan. Choose the company assistant and a plan that includes a data processing contract and lets you decide whether your data is used to train models.
3. Wednesday: a one-page policy. What is allowed, which data never gets pasted, who approves new connections, and that decisions about people are made by people.
4. Thursday: one test connection. Connect a single tool, with minimum access, for one or two people. Check what the assistant can see, what gets logged and how to disconnect it.
5. Friday: one hour of training. Use real cases, or made-up ones where personal data is involved. Note the date, who attended and what you covered: a simple way to document your Article 4 AI literacy measures.

After a month, review which connections are still active, who uses them and whether any can go. If you are also tidying up the rest of your tools, see how to digitize a small business step by step.

Connecting Claude or ChatGPT to UTILIA OS

If your business runs on UTILIA OS, you can connect your Claude or ChatGPT to your company; Cursor and Gemini CLI too. The connectors use OAuth 2.1 with PKCE: each person connects Claude or ChatGPT with their own account and can disconnect it whenever they want. Claude and ChatGPT require a paid plan from those services. We explain it on the for growing page.

UTILIA OS also comes with built-in AI, with an allowance included in every plan; details are on pricing. Before you connect anything, run through the checklist in this guide: decide who connects it, what for and which company data it will look up.