Does AI Train on Your Company Data? How to Check

· DoubleX

Quick answer

It depends on the provider and the plan. Some tools use what you send to improve their models; others contractually do not. Check the provider's terms on training and retention. DoubleX Agentes does not use company content for training and only sends requests to providers that don't either.

Pasting a contract, a customer list, or a spreadsheet into an AI tool raises a fair question: will that content end up shaping a model other people use? The answer varies by tool, plan, and provider, so it pays to know what to look for instead of trusting a marketing page.

What does it mean to train a model on data?#

Training a model means using text, files, or conversations as examples so the model learns from them and changes how it behaves in the future. It is different from simply processing your request: when you ask an AI to summarize a PDF, the model reads the file to produce an answer, and that alone is not training.

Three things get mixed up in privacy discussions:

  • Training: your content is used to improve a model.
  • Retention: your content is stored for some period, for example for abuse monitoring or logs.
  • Access: people or systems that can see your content.

A provider can store data without training on it, and a provider can promise not to train while still keeping logs for a while. You need to check each one separately.

What happens to the content your company sends?#

It depends on where the request goes. In DoubleX Agentes, company content is not used to train models, and requests only go to AI providers that do not use what they receive for training. Each company also has its own database, separate from other companies, and the passwords, tokens, and keys of connected systems are encrypted with AES-256-GCM.

When an agent works, content typically flows through these points:

StepWhat is involvedWhat to verify
You send a requestYour instruction and any filesWho can see the task history
The agent reads a systemEmail, spreadsheet, CRM, documentsWhich actions the agent is allowed to run
The model generates an answerThe provider receives the needed textTraining and retention terms of that provider
The result is savedTask history and filesSharing and permissions inside your company

What is zero data retention and when should you use it?#

Zero data retention is a setting in which the AI provider keeps nothing after it returns the response: no prompt, no output, no stored copy. It goes a step beyond "we don't train on your data," because nothing remains to be reviewed, leaked, or reused later.

It makes the most sense when agents handle:

  • Customer personal data, such as names, tax IDs, and addresses.
  • Financial records, contracts, and pricing.
  • Health, HR, or legal documents.
  • Anything your clients require you to keep out of third-party storage.

DoubleX Agentes offers a zero-retention option on the Models screen: you use only providers that keep nothing after the response. For a deeper look, read Zero Data Retention in AI: What It Is and When to Use It.

How does your choice of models affect privacy?#

Every model you pick means a provider that receives your text, so the choice of model is also a choice of whose terms apply. DoubleX Agentes accesses models by API from labs such as Anthropic, OpenAI, and Google, and you choose which ones to use on the Models screen.

A practical way to decide:

  1. List the agents and the kind of data each one sees.
  2. Mark the ones that handle sensitive information.
  3. For those, restrict the models to providers that meet your retention requirement.
  4. Keep the rest on whatever model gives the best quality for the cost.

Model choice is only one layer. Limiting what an agent can reach matters just as much, and you can set which systems each agent uses and which actions it can run in each one, including read-only access. See How to Control What an AI Agent Can Access and Do.

How do you check the policy of any AI tool?#

Read the terms, then ask the vendor in writing. Public pages often describe consumer plans, while business plans have different rules, so confirm which one applies to you.

Questions to ask any AI vendor:

  • Is my content used to train or improve models? Is that true for every plan?
  • Which model providers receive my data, and do they train on it?
  • How long is data retained, and why (logs, abuse monitoring, support)?
  • Is zero data retention available, and how do I enable it?
  • Is my data stored separately from other customers?
  • How are credentials and tokens for connected systems protected?
  • Can I see a history of what the AI did with my data?
  • How does the vendor handle data protection laws that apply to me?

For the legal side, see AI and Data Privacy Laws: What Businesses Need to Know. DoubleX Agentes follows Brazil's LGPD, and its privacy policy is at doublex.ai/privacidade.

Frequently asked questions#

Does AI always train on the data I type in?#

No. It depends on the provider and the plan. Many business offerings exclude customer content from training, but consumer versions of some tools may not. Always check the terms of the exact plan you use.

Is not training the same as not storing my data?#

No. A provider can skip training and still retain data for a period. Zero data retention is the stronger guarantee, because nothing is kept after the response.

Does DoubleX Agentes train models on my company's content?#

No. Company content is not used to train models, and requests only go to providers that do not use what they receive for training. Each company has its own separate database.

Can I see what an agent did with my data?#

Yes. Each request becomes a task with full history: the request, each tool the agent used, and the result. Before sending, deleting, or changing something in your systems, the agent asks for confirmation.

Do I need zero retention for every agent?#

Not necessarily. Use it for agents that handle personal, financial, legal, or other sensitive data. Agents working with public or low-risk information can use a broader set of models.

Related guides

More Security and privacy guides