---
title: "Attorney-Client Privilege"
description: "Attorney-client privilege protects confidential lawyer-client communications and can be inadvertently waived. Why pasting privileged material into AI tools is a live legal risk."
url: "https://sonomos.ai/glossary/attorney-client-privilege/"
source: "https://sonomos.ai/glossary/attorney-client-privilege/"
---

# Attorney-Client Privilege

A legal doctrine protecting confidential communications between a lawyer and client from compelled disclosure — and one of the easiest privileges to inadvertently waive.

## In depth

Attorney-client privilege protects confidential communications between a lawyer and client made for the purpose of seeking or providing legal advice. The privilege is the client's, not the lawyer's, and is one of the oldest doctrines in common-law systems. Critically for AI users, the privilege can be waived — including by voluntarily disclosing the communication to a third party. Whether pasting a privileged document into a generative AI service constitutes such a disclosure is a developing legal question, and federal courts have begun to address it directly. Even where waiver is uncertain, the prudent posture is not to disclose privileged material to systems whose contents the client cannot control.

## Examples

- A litigator pasting a draft brief that contains privileged client analysis into ChatGPT to get redline suggestions.
- An associate using a coding assistant to refactor a privileged spreadsheet of case strategy notes.
- A paralegal asking an AI tool to summarise privileged email correspondence with the client.

## How Locke handles attorney-client privilege

Locke's Legal industry profile detects patterns associated with privileged material — case numbers, witness identifiers, privilege headers, work-product references. Detection happens locally; nothing leaves the device, which means the act of running Locke itself does not implicate waiver.

[See the product](https://sonomos.ai/locke/)

## Related terms

- Prompt Leakage

  The unintended disclosure of sensitive information through the content of a prompt sent to an AI model.
- Data Masking

  Replacing sensitive values with realistic-looking substitutes so the data remains usable while the originals never leave the device.
- Tokenization

  Replacing a sensitive value with a non-sensitive token that maps back to the original through a separate, controlled lookup.

## Reading on this topic

- [Is Claude Confidential? What Anthropic Can See, Store, and Train On (2026)](https://sonomos.ai/blog/is-claude-confidential-2026/)
- [AI Tools for Legal Research in 2026: Harvey, Westlaw AI, Lexis+ AI, and Why Citation Hallucination Still Matters](https://sonomos.ai/blog/ai-legal-research-harvey-westlaw-casetext-2026/)
- [AI and the Work Product Doctrine: Privilege Risks When Lawyers Use ChatGPT and Claude](https://sonomos.ai/blog/ai-work-product-doctrine-legal-privilege-2026/)
- [Can Lawyers Use ChatGPT? Confidentiality, Privilege, and AI in 2026](https://sonomos.ai/blog/can-lawyers-use-chatgpt-confidentiality-2026/)