Bar Rule 1.6 — Confidentiality and AI Use
When to use this
Apply whenever:
- A lawyer user asks whether they can paste client documents into an AI tool.
- A user asks about data residency, training data policy, or vendor confidentiality.
- A law firm is evaluating AI vendors for client-matter work.
- A user shares raw, unredacted client information in a conversation.
- A compliance review of AI tool use in a law practice is required.
The confidentiality duty — what the rules say
ABA Model Rule 1.6 (US)
"A lawyer shall not reveal information relating to the representation of a client unless the client gives informed consent, the disclosure is impliedly authorized in order to carry out the representation, or the disclosure is permitted" by a listed exception.
The AI implication: transmitting client information to an AI vendor — even as part of a query — is a "revelation" of client information to a third party. Unless the AI vendor has appropriate contractual and technical controls, this could violate Rule 1.6.
MENA analogs
Lebanon: The Code of Professional Conduct (Nizám al-mihna) for Lebanese advocates imposes strict secrecy duties (sirr al-mihna) that prohibit disclosure of client information outside authorized legal proceedings. Using a third-party AI service without appropriate confidentiality safeguards could breach these obligations.
KSA — Saudi Bar: The confidentiality duty is embedded in the Code of Law Practice (Royal Decree M/38) and reinforced by the KSA PDPL — personal data of clients must not be transmitted to processors without a lawful basis and adequate contractual safeguards (Art. 29 PDPL).
UAE — Federal Law on Legal Profession + PDPL: Similar framework; lawyers have professional confidentiality duties plus data-protection obligations under the UAE Federal Decree-Law No. 45 of 2021 on Personal Data Protection.
DIFC / ADGM: DIFC Law No. 5 of 2020 on Data Protection and ADGM Data Protection Regulations 2021 — both GDPR-aligned — impose data-processor contractual requirements equivalent to GDPR Art. 28.
European analogs
UK — SRA Code of Conduct: Principle 6 (behave in a way that maintains the trust the public places in you and the provision of legal services) combined with GDPR data-protection obligations. Law firms are data controllers; AI vendors are data processors requiring a DPA.
France: The secret professionnel under French law (Art. 226-13 Code pénal) is one of the broadest in Europe. The Conseil National des Barreaux has raised concerns about AI tools that process client data outside controlled environments.
EU — GDPR: Lawyers are data controllers; instructing an AI vendor to process client personal data requires a GDPR-compliant DPA (Art. 28), a legal basis for processing, and — for cross-border transfers — standard contractual clauses.
The four AI-specific confidentiality risks
Risk 1: Data transmission to AI vendor
Every query containing client facts is transmitted to the AI vendor's infrastructure. Mitigation: use only AI tools operating under a DPA that prohibits disclosure and use of client data for any purpose other than providing the service.
Risk 2: Training data concern
Consumer AI tools may train on user inputs, potentially incorporating client information into the model's weights and — in the worst case — surfacing it to other users. Mitigation: require a contractual no-training-on-data clause; verify it in the vendor's terms of service or enterprise agreement.
Risk 3: Data residency
Client data may be processed in a jurisdiction with weak privacy protections or government access powers incompatible with confidentiality obligations. Mitigation: specify contractual data residency requirements matching the client's jurisdiction; prefer EU/GDPR-compliant or MENA-region hosting for MENA clients.
Risk 4: Cross-tenant isolation
In shared multi-tenant AI deployments, another law firm's data could theoretically be returned in responses. Mitigation: use platforms with enforced tenant isolation at the storage and retrieval layers — see [[safety-client-confidentiality-cross-tenant]].
Best practices
Technical safeguards
- PII redaction before sending to third-party LLMs — see [[safety-pii-redaction-before-rag]].
- Encryption in transit (TLS 1.3) and at rest (AES-256).
- Audit logs of all AI queries containing client data — accessible to the supervising lawyer.
- Tenant-scoped access: only matter-team members can query matter-related AI history.
Contractual safeguards
- Data Processing Agreement (DPA): mandatory for GDPR/PDPL compliance; covers purpose limitation, sub-processor controls, deletion, incident notification.
- No-training clause: explicit prohibition on training the model on submitted content.
- Data residency clause: specifies which region(s) data may be processed in.
- Confidentiality clause: AI vendor commits to confidentiality obligations.
- Sub-processor list: vendor discloses all sub-processors; notification on changes.
Operational practices
- Update the firm's client engagement letter to disclose AI use and obtain informed consent.
- Never use a consumer-grade AI tool (e.g., a free web-based chatbot without enterprise controls) for matter-specific client data.
- For jurisdictions with heightened sensitivity (KSA, UAE, Lebanon), prefer tools with in-region hosting.
- Document AI tool selection decisions in the firm's technology risk register.
What never to send to AI without redaction
- National ID numbers, passport numbers, civil registration numbers →
[NAT_ID] - IBAN, account, and routing numbers →
[ACCOUNT] - Health information →
[HEALTH] - Client names where the matter is not publicly known →
[CLIENT] - Specific transaction or case details that could identify a party → anonymize/hypothesize
See [[safety-pii-redaction-before-rag]] for the full redaction taxonomy.
MENA-specific data residency considerations
| Jurisdiction | Data-residency preference | Regulatory basis |
|---|---|---|
| KSA | In-region preferred; cross-border transfer requires SDAIA-authorized safeguards | KSA PDPL Art. 29 |
| UAE (onshore) | In-region preferred; contractual safeguards required for cross-border | UAE PDPL Art. 22 |
| DIFC | GDPR-aligned; DIFC Law No. 5/2020 | DIFC DP Law |
| ADGM | GDPR-aligned; ADGM DPR 2021 | ADGM DPR |
| Lebanon | No formal data-protection law as of May 2026; professional secrecy applies | Bar Code of Conduct |
| Egypt | Egypt PDPL (Law 151 of 2020) — transfers permitted with Data Protection Centre approval or adequacy | Egypt PDPL |
Consequences of failure
- Bar discipline: violation of confidentiality is among the most serious professional-conduct offenses in all jurisdictions; sanctions range from reprimand to disbarment.
- Civil liability: clients may sue for breach of confidentiality, professional negligence, or breach of fiduciary duty.
- Regulatory sanctions: data-protection authorities (ICO, CNIL, SDAIA, UAE Data Office) may impose fines under their respective GDPR-equivalent laws.
Related skills
- [[safety-bar-rule-1-1-competence-ai]] — competence duties for AI tool use
- [[safety-bar-rules-confidentiality]] — bar-rules confidentiality architecture
- [[safety-client-confidentiality-cross-tenant]] — cross-tenant isolation guarantees
- [[safety-pii-redaction-before-rag]] — PII redaction before external LLM calls
- [[safety-ai-not-privileged-disclaimer-us-heppner]] — privilege status of AI conversations
- [[safety-cross-border-data-transfer-gcc-eu]] — cross-border data transfer obligations