DeepSeek for an IT Technician: Use Cases, Limits and GDPR

DeepSeek stands out for models available through an API at low cost. Its main issue is data: its privacy policy places processing on servers in China, which led European authorities to react in 2025. Here is what I take from it, what I could not verify, and the usage rule I draw.

Published
Topic
AI
Reading
3 min read
Author
Guillaume Richard

Lire cet article en français

Written on 6 November 2026. The regulatory information dates from 2025; I found no recent source on the status of the proceedings.

My starting point

I tested it, through the API, for development.

What it does well

  • API access: the official documentation presents endpoints compatible with those of OpenAI and Anthropic, which makes integration into existing tools easier.
  • Recent models: the release notes page mentions DeepSeek-V4-Pro and DeepSeek-V4.1-Flash-1017 on the day I read it (no publication date shown).
  • Appeal for development: the documentation mentions integrations with AI-assisted programming tools.

Where I stay wary: data

  • DeepSeek’s privacy policy states, according to the press, that users’ data is stored on servers in China and may be shared with affiliated companies and service providers.
  • In January 2025, the Italian data protection authority (the Garante) blocked access to the app while it checked how data is collected and stored; the Berlin commissioner asked Apple and Google to remove the app, judging the transfer of data to China non-compliant with GDPR; the French CNIL announced it would examine its practices. This information comes from the press of the time.
  • I found no recent source on the current status of these proceedings. Check the statements of the Garante, the Berlin authority and the CNIL before relying on this paragraph.

Practical consequence: avoid it, in its app as on its website, for any professional or personal data.

What interests me: value for money. No client or colleague has asked me about using it in a company. My rule, for every assistant: never a password, bank details or IBAN, and never a first or last name.

IT use cases (without real data)

NeedWhat I ask itPrecaution
Syntax questionA command option, a regular expressionNo client data, nothing internal
Exercise, learningA fictional sample scriptTest in a lab
API integrationEvaluate a model on public or synthetic dataNever production data

Privacy and GDPR

  • Do not paste professional or personal data.
  • For business use, assess the contract and where processing takes place before anything else; if you cannot conclude, pass.
  • If you use published model weights and run them yourself, the data question is different: your own infrastructure then does the processing, and securing it is up to you. Check the licence of the model you use, which I could not confirm on the page I read.

Pricing and versions

I could not read an official price list for the app; the API documentation publishes per-model prices. Check the official platform the day you decide.

What to do

  1. Do not open it to the team for work data without a written assessment.
  2. To test, use fictional data.
  3. Write a rule: “no AI tool that has not been approved for work data”.
  4. Read the authorities’ statements before taking a position.

Sources

Written by

Guillaume Richard
IT & Digital Systems Technician. About

A project, a question?

Write to me: I read every message and answer personally.

Search