FR Newsletter

AI for Work

Using AI at work: confidentiality, internal rules and the EU AI Act basics

Most people now use AI at work, often before their company has said anything about it. Two questions come up: what can I safely give these tools, and what does the new European law actually ask of me? Here are the basics, from the official texts, without the scare stories. This is general information, not legal advice.

First: who does the AI Act apply to?

The EU AI Act (Regulation (EU) 2024/1689) mainly distinguishes two roles. The provider develops an AI system or puts it on the market: the company behind a chatbot, for example. The deployer is any person or organisation using an AI system under its authority, except for purely personal, non-professional use. In practice, when you use an assistant for your job, the deployer is usually your employer or your business, not you personally.

Most obligations in the Act concern providers, and only certain systems. For an office that uses AI to write, translate or summarise, three things matter: AI literacy, transparency, and the list of prohibited practices.

Screenshot coming after our hands-on test
Screenshot of a one-page internal AI policy: coming after our hands-on test.

1. Confidentiality: what not to paste

This part is not new law: it is data protection and business secrecy. The CNIL, France's data protection authority, recommends that users only submit information they are allowed to share, and never confidential information, such as personal data or company data covered by a secret (trade secrets, professional ethics obligations), when using a consumer service.

A simple rule of thumb: if you would not send it to an outside supplier without a contract, do not paste it into a personal AI account. Customer files, salaries, health data, unpublished figures and contracts go only into tools your organisation has approved for that purpose.

2. Write a one-page internal policy

The CNIL recommends that any organisation frame the use of generative AI with internal policies that clearly list authorised and prohibited uses, train users, and carry out regular checks. For a small business, one page is enough:

  • Approved tools, with work accounts rather than personal ones.
  • Data that must never be entered (personal data, confidential documents).
  • Allowed uses (drafting, translation, summaries) and forbidden ones (for example, decisions about people without human review).
  • The check rule: every output is reread by a human before it is used or sent. The CNIL reminds users that these systems can produce wrong results that look plausible.
  • A contact person for questions or incidents.

3. AI literacy (Article 4): already in force

Article 4 has applied since 2 February 2025. It asks providers and deployers of AI systems to take measures on the AI literacy of their staff and of other people who use AI systems on their behalf, taking into account their knowledge, experience and the context of use. The AI Omnibus, which entered into force on 27 July 2026 according to the European Commission, amended this article: literacy remains an obligation for providers and deployers, but no specific or "sufficient" level is now mandated.

What the Commission's Q&A says in practice:

  • A company whose employees use a chatbot to write advertising text or translate should inform them about specific risks, such as hallucinations.
  • There is no one-size-fits-all training and no certificate is needed: an internal record of trainings and guidance is enough.
  • No "AI officer" or governance board is required by Article 4.
  • Supervision is by national market surveillance authorities, from 2 August 2026.

4. Transparency (Article 50): since 2 August 2026

The transparency rules apply from 2 August 2026, as confirmed by the European Commission. They split between providers and deployers:

  • Providers must design chatbots so that people know they are talking to an AI (unless it is obvious), and mark AI-generated audio, image, video or text in a machine-readable way.
  • Deployers must disclose deepfakes (generated or manipulated image, audio or video), and disclose AI-generated or manipulated text published to inform the public on matters of public interest. That text duty does not apply when the content has been reviewed or edited by a human and a person or organisation holds editorial responsibility for it.

If your business puts a chatbot on its website or publishes AI-generated images of real people, these rules concern you. The information must be given clearly, at the latest at the first interaction or exposure.

5. Prohibited practices (Article 5): since 2 February 2025

Some uses of AI are banned outright, whoever does them. The list includes manipulative or deceptive techniques that cause significant harm, exploiting people's vulnerabilities, social scoring, and, directly relevant to employers, AI that infers the emotions of people in the workplace or in education, except for medical or safety reasons. According to the Commission, the AI Omnibus added a ninth ban, on systems generating non-consensual sexual content or child sexual abuse material, which takes effect in December 2026.

6. What is not yet in force

Some AI uses are classed as high-risk, including tools for recruitment and managing workers, such as CV-sorting software. According to the European Commission, following the AI Omnibus, the rules for these uses apply from 2 December 2027, and from 2 August 2028 for AI built into regulated products. If you buy or use such a tool, ask the supplier now how it plans to comply.

What to watch

  • The AI Act does not replace the GDPR. As soon as personal data is involved, data protection rules apply as before.
  • The text keeps moving. The AI Omnibus changed dates and obligations in 2026, and guidelines continue to be published. Check the Commission's pages before relying on a date.
  • This is not legal advice. Your sector, your country and your contracts can add obligations. When in doubt, ask a professional.

Questions

Does my small business have to certify its staff in AI?

No certificate is required, according to the European Commission. Organisations that use AI systems must take measures to support their staff's AI literacy, and can simply keep an internal record of trainings and guidance.

Do I have to say that I used AI to write an email or a report?

The AI Act does not require it for ordinary internal documents. Its labelling duties for users target deepfakes and texts published to inform the public on matters of public interest, with an exception when a human has reviewed the text and someone takes editorial responsibility. Your employer's rules may ask for more.

Is this legal advice?

No. This guide explains the basics from official sources. For your own situation, ask your data protection officer, your legal department or a lawyer.

Sources

  1. Regulation (EU) 2024/1689 (AI Act), Official Journal text, EUR-Lex
  2. European Commission, AI Act (Shaping Europe's digital future), last updated 3 August 2026
  3. European Commission, AI Literacy: Questions and Answers, last updated 27 July 2026
  4. CNIL, Questions and answers on using a generative AI system (in French)