Skip to content
Technology transformation expertise firm · ARDIANE Group Innovate, secure, succeed

Artificial intelligence

What a corporate AI charter must settle, what it leaves to other documents and who signs it

In brief

A corporate AI charter is the document that tells each user which AI tools they may use, which data they must not enter into them, what they must review and whom to report a deviation to. Management Square sets out what this charter must settle, what belongs in other documents (system register, security policy, contracts) and who signs it, without treating it as a sufficient means of compliance.

Page contents

Suppose a civil engineering design office whose executive management commissions a corporate AI charter. Three drafts are circulating. The one from the legal department sets out principles: transparency, fairness, vigilance. The one from the IT department draws up a list of bans. The one from human resources adds a paragraph to the IT charter. None of them tells an engineer whether they can submit a calculation note to the assistant they use at home.

To be useful, an AI acceptable use charter settles four points for the person reading it: the authorised tools, the data that must not be entered, the expected human review and the way to report a deviation. The rest (inventory of systems, security rules, supplier clauses, compliance with data law) belongs in other documents, to which it refers. We propose that it be signed by executive management. This division is a methodological proposal. The charter does not amount to compliance on its own: classifying each use with regard to the texts is a matter for the organisation and its lawyers.

The authorised tools and the data that must not be entered

The list of tools is named and dated. It specifies the account to be used, professional and provided by the business, which rules out using the same tool from a personal account. It covers AI features activated in software already in service and browser extensions, forms described in our entry on shadow AI.

It also says how to request the addition of a tool, to whom, and within what time a response is due. Faced with a closed list with no way to make a request, the user of an unlisted tool may choose to use it without saying so.

The rule on data uses the vocabulary users know: the business’s confidentiality classes, if it has any, and examples drawn from the business itself. For the design office, these are a client’s plans and calculation notes, the prices of a bid in progress, employees’ personal data, passwords, the code of internal tools. The rule intersects with the list of tools: data banned in a consumer assistant may be allowed in a tool whose contract excludes the reuse of inputs.

Two provisions of the GDPR, Regulation (EU) 2016/679, read on 12 November 2026 in its consolidated version, shed light on the case of personal data. Article 9(1) lists special categories, including data concerning health and data revealing political opinions or trade union membership, and prohibits their processing outside the exceptions in paragraph 2: a charter can name them as they stand. Article 29 provides that a person acting under the authority of the controller, who has access to personal data, may only process them on the controller’s instructions, unless required by law. The charter may, in our reading, carry such an instruction; whether it is sufficient is for the data protection officer and the lawyers to say.

Human review and reporting

The charter names the person who answers for content produced with AI: the person who signs it, sends it or incorporates it into a deliverable. It specifies what reviewing means depending on the use: checking each figure and each reference against its source, putting generated code through the usual review, reading a translation intended for a client in full.

It also settles what principles leave open. Should the use of AI be disclosed, and to whom? Which decisions are not taken on the sole basis of the tool’s output, starting with those concerning a person, such as screening a job application?

Reporting covers the mistaken entry of banned data, an incorrect output sent to a client and the discovery of an unlisted tool. For each, the charter gives a recipient and a channel. It also says what the person who reports their own mistake risks, a point to be decided with human resources and the lawyers: without a written answer, the user may prefer to keep quiet.

Article 33(1) of the GDPR requires the controller to notify a personal data breach to the supervisory authority “without undue delay and, where feasible, not later than 72 hours after having become aware of it”, unless it is unlikely to result in a risk to the rights and freedoms of natural persons. Whether a mistaken entry constitutes such a breach is for the organisation and its data protection officer to say; but they still have to find out about it.

What a corporate AI charter leaves to other documents

Our AI governance page lists, among the responses that fail, a charter of principles that nothing links to project decisions. Applicable rules and the references below set this charter apart from it.

Subject Document that carries it What the charter keeps of it
Inventory and classification of AI systems AI system register The way to declare a new tool
Protection of workstations, accounts and the network Security policy, IT charter A reference
Legal basis, record of processing, impact assessments Documentation kept with the data protection officer The rule on entering personal data
Reuse of inputs, place of processing, retention Supplier contracts, purchasing procedure The list of tools, which follows from them
Design of an assistant or agent by a team Project approval process A sentence referring to it

The register, the classification and the approval process are presented on that service page, and how to set them up in our article on implementing AI governance.

The last row marks the limit of the text, written for the user of provided tools. A team that connects an agent to internal applications falls outside its scope, and the rights of that agent are examined according to our article on AI agent security.

Who signs the charter, and who keeps it up to date

Signature by executive management, or by the executive committee member who has authority over all the users concerned, has a practical reason: the text arbitrates between what business units would like to do and what security and legal accept, and it applies to all departments.

Its drafting brings together those who hold each answer: the IT department and the head of security for the tools, the data protection officer and legal for the data, human resources for the scope of the text with regard to employees, and users, who say whether the rules are workable in their job.

Employment law questions arise for the lawyers before distribution, and we do not settle them here. Is the charter enforceable against employees, and on what conditions? How does it fit with the internal rules and the IT charter? Must staff representatives be informed or consulted?

The charter is accompanied by user awareness training, run by the organisation. Regulation (EU) 2024/1689, known as the AI Act, requires in its Article 4, read in the consolidated version of 27 July 2026, that providers and deployers take measures to foster the development of AI literacy among their staff, without requiring them to ensure a specific level of literacy by an individual. Charter and awareness training may, in our reading, count among these measures; they do not exhaust the obligations presented in our article on AI Act obligations for businesses.

A named function keeps the charter up to date, for example the one that runs AI governance. Its application is monitored through a few records: requests to add a tool and their response time, reports received, tools discovered outside the list. The date of the first revision is written into the text from the day it is signed.

Insights

Further reading

All insights

FAQ

Frequently asked questions

Another question? Ask us

Is an AI charter mandatory in a business?

Article 4 of Regulation (EU) 2024/1689 on artificial intelligence, read in its consolidated version of 27 July 2026, does not mention a charter: it asks providers and deployers to take measures to foster the development of AI literacy among their staff. A charter accompanied by awareness training may, in our reading, count among these measures, without summing them up. Saying what is binding on a given organisation, under this text or another, is a matter for its lawyers.

What is the difference between an AI charter and AI governance?

The charter is addressed to each user and tells them which tools to use, which data not to enter, what to review and whom to report a deviation to. AI governance is addressed to the organisation: it lists the systems, classifies them by risk level, appoints their owners and sets up an approval process. The charter is one of the documents of this governance, the one that users read.

Should the AI charter ban consumer AI assistants?

That is a choice the charter must write down, whatever it is. It can ban them, allow them for content with no confidential or personal data, or replace them with a tool under contract. A ban with no alternative leaves users' needs unmet: the charter benefits from stating, next to each ban, the authorised tool for the same task.

Does the AI charter apply to contractors and temporary staff?

The answer depends on contracts and employment law, and is a question for the organisation's lawyers. Article 4 of Regulation (EU) 2024/1689 covers, in addition to staff, "other persons dealing with the operation and use of AI systems on their behalf". It is therefore prudent to plan, from the drafting stage, how the text is brought to the attention of external contributors.

When should an AI charter be revised?

We propose two triggers. The first is a date, written into the charter on the day it is signed. The second is an event: a tool added or withdrawn, a change in a supplier's terms, a reported incident, a change in the applicable texts. Placing the list of tools in an annex can avoid a new signature of the main text at each update, subject to the formalities the organisation's lawyers have indicated.

A project to bring to completion?

A first conversation is enough to understand your challenge, identify the right people and tell you frankly what we can bring.