---
id: CP-314
slug: company-ai-use-policy
lang: en
title: "Company AI use policy: allow, restrict, prohibit"
description: What a company AI use policy is for, what it contains, and what you do before opening a new generative AI tool.
canonical_url: https://startbrain.io/cyberpedia/en/articles/company-ai-use-policy/
markdown_url: https://startbrain.io/cyberpedia/en/articles/company-ai-use-policy.md
publisher: Brain (https://startbrain.io)
license: CC BY 4.0 (https://creativecommons.org/licenses/by/4.0/)
cite_as: "Brain, \"Company AI use policy: allow, restrict, prohibit\", Cyberpedia, 2026. https://startbrain.io/cyberpedia/en/articles/company-ai-use-policy/ (CC BY 4.0)"
wing: 4
theme: authorized_tools
reflex: air-010
schema: G5
updated: "2026-11-14"
translations:
  en: https://startbrain.io/cyberpedia/en/articles/company-ai-use-policy/
  fr: https://startbrain.io/cyberpedia/fr/articles/charte-utilisation-ia-entreprise/
  de: https://startbrain.io/cyberpedia/de/articles/ki-nutzungsrichtlinie-unternehmen/
  es: https://startbrain.io/cyberpedia/es/articles/politica-interna-uso-ia-empresa/
  it: https://startbrain.io/cyberpedia/it/articles/regolamento-utilizzo-ia-azienda/
  nl: https://startbrain.io/cyberpedia/nl/articles/ai-gebruiksreglement-onderneming/
  pl: https://startbrain.io/cyberpedia/pl/articles/regulamin-korzystania-z-ai-w-firmie/
  pt: https://startbrain.io/cyberpedia/pt/articles/carta-utilizacao-ia-empresa/
  ja: https://startbrain.io/cyberpedia/ja/articles/company-ai-use-policy/
sources:
  - "n": 1
    title: Comment déployer une IA générative ? La CNIL apporte de premières précisions
    publisher: CNIL
    url: https://www.cnil.fr/fr/comment-deployer-une-ia-generative-la-cnil-apporte-de-premieres-precisions
    published: "2024-07-18"
    accessed: "2026-11-14"
  - "n": 2
    title: Les questions-réponses de la CNIL sur l’utilisation d’un système d’IA générative
    publisher: CNIL
    url: https://www.cnil.fr/fr/les-questions-reponses-de-la-cnil-sur-lutilisation-dun-systeme-dia-generative
    published: "2024-07-18"
    accessed: "2026-11-14"
  - "n": 3
    title: Security recommendations for a generative AI system
    publisher: ANSSI
    url: https://messervices.cyber.gouv.fr/documents-guides/security_recommandations_for_a_generative_ai_system.pdf
    published: "2024-04-29"
    accessed: "2026-11-14"
  - "n": 4
    title: AI Literacy - Questions & Answers
    publisher: European Commission
    url: https://digital-strategy.ec.europa.eu/en/faqs/ai-literacy-questions-answers
    accessed: "2026-11-14"
  - "n": 5
    title: "ChatGPT and large language models: what's the risk?"
    publisher: NCSC (UK)
    url: https://www.ncsc.gov.uk/blog-post/chatgpt-and-large-language-models-whats-the-risk
    published: "2023-03-14"
    accessed: "2026-11-14"
  - "n": 6
    title: AI Risk Management Framework
    publisher: NIST
    url: https://www.nist.gov/itl/ai-risk-management-framework
    published: "2023-01-26"
    accessed: "2026-11-14"
  - "n": 7
    title: "AI and cyber security: what you need to know"
    publisher: NCSC (UK)
    url: https://www.ncsc.gov.uk/guidance/ai-and-cyber-security-what-you-need-to-know
    published: "2024-02-13"
    accessed: "2026-11-14"
  - "n": 8
    title: LLM02:2025 Sensitive Information Disclosure
    publisher: OWASP
    url: https://genai.owasp.org/llmrisk/llm022025-sensitive-information-disclosure/
    accessed: "2026-11-14"
---

# Company AI use policy: allow, restrict, prohibit

An AI use policy is an internal document that tells employees which uses of generative AI are allowed and which are ruled out. The CNIL recommends restricting uses by defining a list of authorized and prohibited uses [1].

This entry sorts those rules into three columns: what is allowed with approved tools, what needs authorization, and what is prohibited. It also says what you do before using a new tool and who to ask your questions.

## In 30 seconds

- Where it happens: whenever an employee uses generative AI to produce content, rework a text (correction, translation) or analyze documents (sorting, summarizing) [2].
- The risk: according to ANSSI, with most online services, the data sent is collected and used by the provider to optimize its models [3].
- The reflex: read the policy before using a new tool, and enter no sensitive data into a public online AI service [3]; if something goes wrong, go through the contact point the organization may have designated for raising ethical difficulties [2].

## Straight answers

### Why have a policy?

According to the CNIL, in the general case it is the user organization that bears legal responsibility if its staff misuse AI [2]. It therefore recommends that every organization restrict use through internal policies or charters that clearly define authorized and prohibited uses [2].

### What does a policy contain?

As examples of prohibitions, the CNIL gives not providing personal data to the system and not entrusting it with decision-making [1]. Depending on the deployment mode, it will also be advisable to prohibit providing certain confidential data, for example data covered by industrial and commercial secrecy [2]. It recommends training users appropriately and carrying out regular checks [2].

### Who takes part in drafting it?

The CNIL recommends involving all stakeholders from the start: the data protection officer (DPO), the head of information systems, the CISO and the business managers [1]. The DPO can usefully play a governance role, where appropriate together with the CISO, and this designation can give users a contact point for raising their problems [2].

### Does the EU AI Act require employee training?

The European Commission explains that Article 4 of the AI Act requires providers and deployers of AI systems to take measures to promote the AI literacy of their staff and of other people who use these systems on their behalf [4]. These measures take into account the technical knowledge, experience, education and training of those people [4].

The FAQ specifies that Article 4 does not require measuring employees' knowledge, nor guaranteeing a specific level of literacy for each person [4]. It also states that no certificate is needed (the organization may keep an internal record of training) and that no particular governance structure is required [4]. It sets no mandatory training format, but considers that relying on instructions for use, or asking staff to read them, may be ineffective [4].

According to the same FAQ, Article 4 became applicable on 2 February 2025 [4]. The supervision and penalty rules have applied since 2 August 2026, and supervision falls to the national market surveillance authorities [4]. The Digital Omnibus on AI, which entered into force in mid-July 2026, amended Article 4: the AI literacy obligation remains for providers and deployers, but no specific or sufficient level is imposed [4].

### Is a company whose employees use ChatGPT to write or translate concerned?

The Commission answers yes: these employees should be informed of the risks specific to the tool, for example hallucination [4].

### Can I enter company data into a public AI service?

ANSSI writes that sensitive data must absolutely not be sent to third-party generative AI services such as ChatGPT, Gemini, Copilot, DeepL or Perplexity [3]. It counts among this data passwords and API keys, personal data, and the company's contractual, legal and financial data [3]. The NCSC (UK) recommends not including sensitive information in queries to public LLMs, and not submitting queries that would cause problems if they were made public [5].

## How it works

The CNIL advises starting from a concrete need rather than deploying generative AI with no precise purpose [1]. The deployment mode matters: for non-confidential uses, a consumer service can be considered with appropriate safeguards. If the use involves providing personal data or sensitive documentation, the CNIL says it generally seems more appropriate and safer to favor an on-premises solution, which limits the risks of data extraction by a third party [2]. With an API, control of the system lies almost entirely in the provider's hands, hence the instruction to avoid entering personal data there as far as possible [2].

Frameworks help organizations assess these risks. NIST's AI risk management framework is intended for voluntary use [6]. Its generative AI profile, published on 26 July 2024, can help organizations identify the risks specific to generative AI [6].

On the threat side, the NCSC (UK) describes prompt injection, where an attacker creates an input designed to make the model behave in an unintended way, for example by revealing confidential information [7]. OWASP lists sensitive information disclosure among LLM risks: personal data, financial data, health records, confidential business data, security credentials and legal documents [8].

## What you do

### Allowed with approved tools

These are the uses the policy lists, on the tools the organization has chosen. The CNIL distinguishes three categories of tasks: generating content, reprocessing existing content (correction, translation) and analyzing data (sorting, summarizing documents) [2]. It gives an example: making a chatbot available to employees to help with writing carries less risk than using a system to help decide about customers, candidates or citizens [2].

Within this framework, the CNIL states that users should submit only information they are authorized to share, and take a critical look at outputs: accuracy, plagiarism, bias [2]. In its view, a good practice is never to reproduce the system's outputs as they are [2].

### Needs authorization

The CNIL considers a consumer service for non-confidential uses, with appropriate safeguards: avoid accounts created with personal addresses and, where applicable, disable the provider's ability to reuse usage data [2]. When the data risks being reused by the provider, the organization carries out a case-by-case analysis to decide whether it prohibits all personal data or only certain categories [2].

When a tool connects to email, documents or code repositories, ANSSI asks that its access rights be reviewed on activation, then regularly (for example every month) [3].

### Prohibited

The policy may prohibit, for example, providing personal data to the system and entrusting it with decision-making [1]. ANSSI, for its part, writes that sensitive data must not be sent to third-party generative AI services [3]. The CNIL cites, among the data to protect when using a consumer service, data covered by business secrecy or by professional ethics obligations [2]. For code, ANSSI strongly recommends not having an AI generate blocks of code intended for critical modules: cryptography, access rights management, sensitive data processing [3].

### Before using a new tool

- Check that it appears in the policy and that the intended use is allowed there.
- Assume that the provider reuses the data entered: according to the NCSC, queries are stored and almost certainly used, at some point, to develop the service or the model, hence the need to understand the terms of use and the privacy policy before asking a sensitive question [5].
- Use a work account rather than a personal address [2].
- If you have a question or a problem, go to the contact point the organization has designated, if there is one: the CNIL states that designating a DPO, who can usefully play a governance role, where appropriate together with the CISO, can give users a contact point for raising their problems [2].

## Pitfalls

A plausible answer is not necessarily accurate: the CNIL notes that these systems can generate inaccurate results that look plausible, and that excessive trust without verification can lead to wrong decisions [2]. It also warns about automation bias, that is, the loss of trust or skill linked to excessive use [2].

A question can be sensitive because of who asks it. The NCSC (UK) gives the example of an executive who asks how to dismiss an employee, and points out that queries made with the same account can be aggregated [5].

An instruction given to the tool does not replace the rule. OWASP notes that restrictions placed in the system prompt may not be followed and may be bypassed through prompt injection [8].

Finally, the burden should not fall on the user alone: the NCSC (UK) writes that customers generally lack the expertise to deal with AI-related risks on their own, and that developers should take responsibility for their customers' security [7].

## Sources

1. CNIL, "Comment déployer une IA générative ? La CNIL apporte de premières précisions", 18 July 2024. https://www.cnil.fr/fr/comment-deployer-une-ia-generative-la-cnil-apporte-de-premieres-precisions (accessed 14 November 2026)
2. CNIL, "Les questions-réponses de la CNIL sur l’utilisation d’un système d’IA générative", 18 July 2024. https://www.cnil.fr/fr/les-questions-reponses-de-la-cnil-sur-lutilisation-dun-systeme-dia-generative (accessed 14 November 2026)
3. ANSSI, "Security recommendations for a generative AI system", 29 April 2024. https://messervices.cyber.gouv.fr/documents-guides/security_recommandations_for_a_generative_ai_system.pdf (accessed 14 November 2026)
4. European Commission, "AI Literacy - Questions & Answers". https://digital-strategy.ec.europa.eu/en/faqs/ai-literacy-questions-answers (accessed 14 November 2026)
5. NCSC (UK), "ChatGPT and large language models: what's the risk?", 14 March 2023. https://www.ncsc.gov.uk/blog-post/chatgpt-and-large-language-models-whats-the-risk (accessed 14 November 2026)
6. NIST, "AI Risk Management Framework", 26 January 2023. https://www.nist.gov/itl/ai-risk-management-framework (accessed 14 November 2026)
7. NCSC (UK), "AI and cyber security: what you need to know", 13 February 2024. https://www.ncsc.gov.uk/guidance/ai-and-cyber-security-what-you-need-to-know (accessed 14 November 2026)
8. OWASP, "LLM02:2025 Sensitive Information Disclosure". https://genai.owasp.org/llmrisk/llm022025-sensitive-information-disclosure/ (accessed 14 November 2026)

---

Original source: Cyberpedia, the public cybersecurity and AI encyclopedia published by Brain. Canonical version: https://startbrain.io/cyberpedia/en/articles/company-ai-use-policy/
You may reuse this text freely under the CC BY 4.0 license (https://creativecommons.org/licenses/by/4.0/), provided you credit "Brain, Cyberpedia" with a link to the canonical URL above.
