Before sending any audited document to an AI tool, take three steps without exception: pseudonymize sensitive data consistently, verify that your Data Protection Impact Assessment (DPIA) covers this processing, and sign a data processing agreement (DPA) that includes a zero-entrainment clause. Document each step. A listener will not be satisfied with your word.
- Classify first: distinguish confidential data (contracts, balance sheets), personal data (names, emails, contract numbers) and public data before any processing.
- Pseudonymize consistently: replace “Jean Dupont” with “Client_A” throughout the document, not just at the first occurrence.
- Encrypt in transit: use TLS 1.3 or equivalent between your workstation and the AI tool API.
- Log: keep access logs, proof of pseudonymization and, if you use a third-party tool, proof of data deletion after processing.
Pro tip: Always check that your DPA contains an explicit clause prohibiting the vendor from using your data to train its public models. Without this guarantee, your documents may feed into models accessible to third parties.
Table of contents
- What GDPR obligations apply when you use AI on audited data?
- Pseudonymization or anonymization: how to choose according to your audit context?
- How to pseudonymize your documents before sending them to an AI?
- What an auditor actually expects from your compliance file
- What technical risks does AI introduce and how can they be contained?
- Key Points
- Why pseudonymization is a real game-changer in audits
- Safe-doc: the layer of protection for your AI-audited documents
- Official references to keep in your audit file
What GDPR obligations apply when you use AI on audited data?
The CNIL recommends an AIPD for any AI system processing personal data in volume or involving automated decision-making. For companies subject to audit, this obligation is not optional: article 35.7 of the GDPR imposes a precise minimum content.
A valid AIPD must cover:
- The description of the processing operations and their purposes.
- The assessment of proportionality: are the data processed strictly necessary?
- Risk assessment for the people concerned (flight, reidentification, discrimination).
- The planned security measures: pseudonymization, encryption, access control.
Beyond the AIPD, there are three related obligations. The processing register must mention each AI use. Any service provider processing data on your behalf must sign a DPA in accordance with Article 28 of the GDPR. In the event of a violation, article 33 requires notification to the CNIL within 72 hours.
The CNIL 2024 security guide specifies that governance, flow mapping and periodic audits are part of the minimum documentary expectations. An auditor who does not find these elements in your file will consider that the treatment is not controlled.
Pseudonymization or anonymization: how to choose according to your audit context?
The distinction is legally structuring. Anonymized data can no longer be linked to a natural person by any reasonably available means: it falls outside the scope of the GDPR. pseudonymised according to article 4(5) data remains personal data, but with a reinforced level of protection.
For an audit context, controlled reversible pseudonymization is often preferable to full anonymization, precisely because the auditor may need to trace information back to its source. Irreversible anonymization removes this possibility.
Practical selection criteria:
- HR data (pay slips, evaluations): reversible pseudonymization if the DPO or internal auditor must be able to remove the mask; anonymization if the data is used only for aggregate statistics.
- Sensitive contracts: consistent pseudonymization of parties, amounts and references - the AI can analyze the structure without exposing the identity of the co-contractors.
- Financial reports: masking of account numbers and IBAN, replacement of entity names with generic identifiers (Entity_1, Entity_2).
Pro tip: If you opt for reversible pseudonymization, the correspondence table (which associates “Client_A” with “Jean Dupont”) must be stored separately, encrypted, and accessible only to authorized people. Its compromise voids all protection.
How to pseudonymize your documents before sending them to an AI?
Effective pseudonymization is not about removing a name: it requires consistent replacements across the entire document and careful management of lookup tables.
1. Classify the document: identify sensitive fields (surnames, first names, emails, contract numbers, financial extracts, IP addresses, SIRET numbers).
2. Define the replacement rules: each entity receives a stable generic identifier (“Supplier_B”, “Contract_2024_003”). The same name must always produce the same identifier in the document.
3. Apply pseudonymization: via an automated script, a preprocessing module or a dedicated solution. Verify that the file metadata (author, revision history) is also purged.
4. Test non-identifiability: subject the pseudonymized document to a manual or automated re-identification attempt. If an external reader can find the real identity from the context, pseudonymization is insufficient.
5. Encrypt and transmit: Send the document over an encrypted connection (TLS 1.3). Keep a time-stamped log of each shipment.
6. Log and archive: document who pseudonymized what, when, and with what tool. This traceability is the proof that the auditor will ask for.
| Sensitive field | Recommended method | Example of replacement |
|---|---|---|
| Last name / first name | Consistent replacement | Jean Dupont → Client_A |
| Tokenization | j.dupont@acme.fr → email_anon@anon.local | |
| Contract number | Generic identifier | CTR-2024-0042 → Contract_2024_003 |
| Financial amount | Partial masking or generalization | - |
| IBAN / RIB | Deletion or tokenization | FR76... → IBAN_anon |
Pro tip: For confidential accounting data, automate pseudonymization upon entry into your AI feed. Manual processing is a source of inconsistencies that the auditor will detect immediately.
What an auditor actually expects from your compliance file
An audit is more than just checking that you have signed a DPA. The CNIL 2024 guide is explicit: governance, flow mapping and documentary evidence form an inseparable whole.
Documentary evidence expected:
- Up-to-date processing register, mentioning each AI use.
- AIPD finalized with action plan followed.
- Time-stamped access logs and evidence of data deletion after processing.
- DPA signed with each AI service provider, including “zero training” clause and deletion deadline.
Governance to be formalized:
1. Appoint a DPO and a CISO with direct access to management.
2. Mapping data flows: who sends what, to which tool, from which station.
3. Keep a register of AI uses: authorized tools, prohibited tools, risk linked to Shadow AI reporting procedure.
4. Train teams: an annual session on AI risks and data protection is rarely enough; favor targeted reminders each time a new tool is deployed.
Regarding contractual clauses, the CNIL checklist is unambiguous: the “zero training” clause, the contractual audit right and the explicit data deletion policy are proof of control that auditors systematically request. Add data localization: a sovereign accommodation in France helps to avoid the risks linked to extraterritorial legislation such as the American Cloud Act.
What technical risks does AI introduce and how can they be contained?
AI-specific threats go beyond traditional vulnerabilities. Three deserve particular attention in an audit context.
- Data poisoning: A malicious actor modifies training data to bias model results. integrity validation by HMAC and dataset version logging can detect these alterations.
- Prompt injection: a malicious user inserts instructions into the document to manipulate the AI response. Filtering and cleaning entries before sending reduces this risk.
- Shadow AI: employees use unapproved public AI tools with sensitive documents. Providing a secure alternative with integrated pseudonymization is more effective than a pure ban, which only pushes uses into the shadows.
Technical controls to be put in place:
- Encryption in transit (TLS 1.3) and at rest (AES-256 or equivalent).
- Granular access control (RBAC): only authorized people access data and models.
- Audit logs analyzed regularly, with alerts on abnormal access.
- Require providers to have ISO 27001, HDS or SOC 2 certification, and an explicit data localization policy.
Pro tip: When selecting an AI tool, always ask for the list of subcontractors and their locations. A tool hosted in Europe can use American infrastructure for certain operations - which is enough to invalidate a sovereignty clause.
Key points
The protection of data from audited companies using AI is based on three inseparable pillars: consistent pseudonymization before any sending, documented AIPD, and DPA with “zero training” clause.
| Point | Details |
|---|---|
| Consistent pseudonymization | Replace each identifier with a stable alias throughout the document before sending to AI. |
| Mandatory AIPD | Any high-risk AI processing requires DPIA in accordance with Article 35.7 of the GDPR. |
| Critical contractual clauses | The DPA must include “zero training” clause, right of audit and deadline for deletion. |
| Sovereign Hosting | Choose a French host to avoid the risks linked to extraterritorial legislation. |
| Safe-doc | Safe-doc pseudonymizes your documents in real time, without storage, with integrated DPA and GDPR compliance. |
Why pseudonymization is a real game-changer in audits
What is striking in the audit files that we see circulating is less the absence of technical measures than the absence of evidence. The teams sometimes pseudonymize, often encrypt, but almost never document. But an auditor can only attest to what he can verify.
Documentary pseudonymization has an advantage that is underestimated: it forces teams to think about their data before processing it. This moment of classification, even brief, mechanically reduces the Shadow AI. When a collaborator knows that he must pseudonymize before sending, he uses the approved tool rather than the public application which does not ask anything. Training is not a barrier to productivity: it’s what makes adoption sustainable.
The real leverage is the combination of AIPD + pseudonymization pipeline + logging. Together, they produce an audit file that the DPO can defend without improvising. Separately, each of these elements remains fragile.
Safe-doc: the layer of protection for your AI-audited documents
Your teams already use ChatGPT, Claude or other AI tools. Safe-doc is inserted into this flow without modifying it: it pseudonymizes your sensitive documents in real time, before sending, without ever storing the files. For legal departments, teams in charge of financial audits or M&A due diligence operations, this is the difference between a solid compliance file and uncontrolled exposure.
Safe-doc provides a signed DPA, a verifiable “zero storage” architecture, and hosting under French jurisdiction. The auditors find in the file what they are looking for: proof of pseudonymization, access logs, contractualized “zero training” clause. Consult DPO and compliance page to check contractual guarantees and start your compliance.
Official references to keep in your audit file
These sources are authoritative among auditors and French supervisory authorities. Include them in your compliance file.
- Security: manage data security | CNIL: reference for governance, DPD/RSSI appointment and periodic audits.
- AI: guaranteeing the security of the development of an AI system | CNIL: technical measures recommended by the CNIL for AI systems (logging, encryption, access control, pseudonymization).
- Personal data security guide 2024 | CNIL: reference document for audits; covers governance, mapping and technical measures.
- Personal data security checklist | CNIL: operational checklist to attach to the audit file.
- Security: risk analysis | CNIL: AIPD methodology and mandatory criteria for AI processing.
Keep these links in your processing register and update them with each annual review of your data protection policy.
This article is general information. For your specific situation, consult your DPO or a lawyer specializing in data protection, and check the texts in force with the CNIL.