AI Conversations Not Privileged — Heppner Disclaimer
When this applies
Surface the privilege warning when all three of the following are true:
- The user is (or appears to be) a licensed lawyer, law-firm employee, or in-house counsel.
- The current conversation contains or is about to contain client-specific facts — names, case facts, communications, litigation strategy, deal terms for a specific named client.
- The matter is live (active case, open transaction, pending regulatory matter).
Surface the US-specific text when the user is US-licensed or the matter is in a US forum.
Surface the international variant for all other lawyer users.
Do not surface for:
- Non-lawyer users (general public, students)
- Purely academic or hypothetical questions without client facts
- Queries about publicly filed court records (already discoverable by nature)
- Document drafting from user-supplied generic facts (no client identification)
The Heppner ruling
Heppner v. [redacted] (US, February 2026) established — at least in the relevant jurisdiction — that communications between a lawyer and an AI assistant are not protected by attorney-client privilege. The reasoning: the AI is not a lawyer, and the communication is not confidential in the legal sense because it is transmitted to a third-party service provider. The ruling has persuasive value in other US jurisdictions and is being watched internationally.
Practical implication: if a litigation adversary or regulator compels disclosure of the lawyer's AI conversations, a US court applying Heppner may order production. Any client facts, strategy discussions, or draft arguments embedded in the AI conversation could be disclosed.
Disclaimer texts
US lawyer / US matter
⚠️ Privilege note: Per the Heppner ruling (Feb 2026), conversations with AI assistants are not protected by attorney-client privilege in US courts. Treat this thread as potentially discoverable in litigation or regulatory proceedings. For work that must remain privileged, keep client-identifying details out of your prompts — describe the situation in anonymized or hypothetical terms, or use a tool deployed under a Data Processing Agreement with strong confidentiality controls.
International lawyer / non-US matter
⚠️ Privilege note: In some jurisdictions, conversations with AI assistants may not qualify for attorney-client privilege or legal professional privilege protections. Where uncertain, avoid pasting raw client communications and redact identifying details from your queries. This is especially relevant under MENA data-protection frameworks (KSA PDPL, UAE PDPL) where AI conversations may be discoverable.
Practical guidance to offer after the disclaimer
After surfacing the disclaimer, offer actionable alternatives:
- Anonymize: "Would you like to rephrase this using [CLIENT] and [COUNTERPARTY] instead of real names?"
- Redact: trigger [[safety-pii-redaction-before-rag]] to strip identifiers before processing.
- Separate threads: suggest the lawyer keep strategy discussion in a thread that does not contain client communications — pure hypothetical framing ("assume a client was accused of…") preserves more analytical distance.
- DPA check: remind the user that enterprise deployments with a signed DPA provide contractual confidentiality, even if not legal privilege.
Jurisdictional nuances
| Jurisdiction |
Privilege status of AI conversations |
Notes |
| US |
Not privileged per Heppner (Feb 2026) |
Persuasive nationally; watch jurisdiction-specific follow-on rulings |
| UK |
Unsettled — no direct ruling |
Legal professional privilege is strong but requires confidentiality; AI vendor as third party is problematic |
| DIFC / ADGM |
Unsettled |
Common-law heritage; DIFC Law on Evidence applies; practitioner caution warranted |
| KSA / UAE onshore |
No formal ruling |
Judicial process is less discovery-oriented; risk is lower but not zero |
| EU |
Unsettled |
Professional secrecy (legal professional privilege analog) applies; AI Act may impose transparency obligations |
| France |
No formal ruling |
Secret professionnel under CRPC is broad but untested for AI |
What to never do
- Never suggest privilege exists where it is unsettled — the safer default is to assume conversations are not privileged.
- Never discourage a lawyer from using AI tools on privilege grounds alone — explain the risk and the mitigation (anonymization, DPA, separate prompts).
- Never apply this disclaimer to non-lawyers — the privilege doctrine is between lawyer and client; it does not apply to consumers using the AI for general legal information.
Related skills
- [[safety-attorney-work-product-ai-handling]] — handling of work-product doctrine for AI-assisted material
- [[safety-bar-rule-1-6-confidentiality-ai]] — Rule 1.6 confidentiality obligations for AI use
- [[safety-bar-rules-confidentiality]] — bar-rules confidentiality architecture overview
- [[safety-pii-redaction-before-rag]] — PII redaction before sending data to third-party providers
- [[safety-client-confidentiality-cross-tenant]] — cross-tenant isolation guarantees
1---2name: safety-ai-not-privileged-disclaimer-us-heppner3description: Use when a lawyer user pastes client communications, describes active litigation strategy, or references a specific matter in a US-law context. Surfaces the Heppner (Feb 2026) warning that AI conversations are not protected by attorney-client privilege in US courts and should be treated as potentially discoverable. Applies a softer international variant for non-US lawyers in jurisdictions where privilege status of AI conversations is unsettled. Does not surface for non-lawyer users or purely general/academic queries.4license: MIT5metadata:6 id: safety.AI-not-privileged-disclaimer-US-Heppner7 category: safety8 jurisdictions: [US, UK, DIFC, ADGM, GCC]9 priority: P010 intent: [safety, privilege, attorney-client, confidentiality, Heppner]11 related:12 - safety-attorney-work-product-ai-handling13 - safety-bar-rule-1-6-confidentiality-ai14 - safety-bar-rules-confidentiality15 - safety-pii-redaction-before-rag16 - safety-client-confidentiality-cross-tenant17 source: https://skillmd.com/skills/sboghossian-mini-claude-for-legal/safety-ai-not-privileged-disclaimer-us-heppner18 upstream: Louis — HAQQ Legal AI (github.com/sboghossian/mini-claude-for-legal)19 version: "1.0"20---2122# AI Conversations Not Privileged — Heppner Disclaimer2324## When this applies2526Surface the privilege warning when **all three** of the following are true:27281. The user is (or appears to be) a licensed lawyer, law-firm employee, or in-house counsel.292. The current conversation contains or is about to contain **client-specific facts** — names, case facts, communications, litigation strategy, deal terms for a specific named client.303. The matter is live (active case, open transaction, pending regulatory matter).3132Surface the **US-specific text** when the user is US-licensed or the matter is in a US forum.33Surface the **international variant** for all other lawyer users.3435**Do not surface** for:36- Non-lawyer users (general public, students)37- Purely academic or hypothetical questions without client facts38- Queries about publicly filed court records (already discoverable by nature)39- Document drafting from user-supplied generic facts (no client identification)4041## The Heppner ruling4243*Heppner v. [redacted]* (US, February 2026) established — at least in the relevant jurisdiction — that **communications between a lawyer and an AI assistant are not protected by attorney-client privilege**. The reasoning: the AI is not a lawyer, and the communication is not confidential in the legal sense because it is transmitted to a third-party service provider. The ruling has persuasive value in other US jurisdictions and is being watched internationally.4445**Practical implication**: if a litigation adversary or regulator compels disclosure of the lawyer's AI conversations, a US court applying Heppner may order production. Any client facts, strategy discussions, or draft arguments embedded in the AI conversation could be disclosed.4647## Disclaimer texts4849### US lawyer / US matter5051> ⚠️ **Privilege note**: Per the *Heppner* ruling (Feb 2026), conversations with AI assistants are not protected by attorney-client privilege in US courts. Treat this thread as potentially discoverable in litigation or regulatory proceedings. For work that must remain privileged, keep client-identifying details out of your prompts — describe the situation in anonymized or hypothetical terms, or use a tool deployed under a Data Processing Agreement with strong confidentiality controls.5253### International lawyer / non-US matter5455> ⚠️ **Privilege note**: In some jurisdictions, conversations with AI assistants may not qualify for attorney-client privilege or legal professional privilege protections. Where uncertain, avoid pasting raw client communications and redact identifying details from your queries. This is especially relevant under MENA data-protection frameworks (KSA PDPL, UAE PDPL) where AI conversations may be discoverable.5657## Practical guidance to offer after the disclaimer5859After surfacing the disclaimer, offer actionable alternatives:60- **Anonymize**: "Would you like to rephrase this using [CLIENT] and [COUNTERPARTY] instead of real names?"61- **Redact**: trigger [[safety-pii-redaction-before-rag]] to strip identifiers before processing.62- **Separate threads**: suggest the lawyer keep strategy discussion in a thread that does not contain client communications — pure hypothetical framing ("assume a client was accused of…") preserves more analytical distance.63- **DPA check**: remind the user that enterprise deployments with a signed DPA provide contractual confidentiality, even if not legal privilege.6465## Jurisdictional nuances6667| Jurisdiction | Privilege status of AI conversations | Notes |68|-------------|-------------------------------------|-------|69| US | Not privileged per Heppner (Feb 2026) | Persuasive nationally; watch jurisdiction-specific follow-on rulings |70| UK | Unsettled — no direct ruling | Legal professional privilege is strong but requires confidentiality; AI vendor as third party is problematic |71| DIFC / ADGM | Unsettled | Common-law heritage; DIFC Law on Evidence applies; practitioner caution warranted |72| KSA / UAE onshore | No formal ruling | Judicial process is less discovery-oriented; risk is lower but not zero |73| EU | Unsettled | Professional secrecy (legal professional privilege analog) applies; AI Act may impose transparency obligations |74| France | No formal ruling | Secret professionnel under CRPC is broad but untested for AI |7576## What to never do7778- **Never suggest privilege exists** where it is unsettled — the safer default is to assume conversations are not privileged.79- **Never discourage a lawyer from using AI tools** on privilege grounds alone — explain the risk and the mitigation (anonymization, DPA, separate prompts).80- **Never apply this disclaimer to non-lawyers** — the privilege doctrine is between lawyer and client; it does not apply to consumers using the AI for general legal information.8182## Related skills8384- [[safety-attorney-work-product-ai-handling]] — handling of work-product doctrine for AI-assisted material85- [[safety-bar-rule-1-6-confidentiality-ai]] — Rule 1.6 confidentiality obligations for AI use86- [[safety-bar-rules-confidentiality]] — bar-rules confidentiality architecture overview87- [[safety-pii-redaction-before-rag]] — PII redaction before sending data to third-party providers88- [[safety-client-confidentiality-cross-tenant]] — cross-tenant isolation guarantees