If a client (or potentially a lawyer) runs facts, theories, timelines, or “talking points for counsel” through a third-party AI tool...

2025-11-28 · Source: Pascal’s Substack · Field: Legal & Regulatory — Legal Technology (LegalTech), Compliance & Risk Management, Litigation & Dispute Resolution · Depth: Intermediate, medium

Summary

A recent bench ruling by Judge Jed Rakoff in the Southern District of New York determined that documents generated by a defendant using a commercial AI tool (Claude by Anthropic) and subsequently shared with legal counsel were not protected by attorney-client privilege or work-product doctrine. The government argued that the AI tool is not a lawyer, the interaction was not "for legal advice" in a legally relevant way, the prompts and outputs were not confidential due to third-party platform involvement, and privilege cannot be "laundered" after the fact. This ruling highlights a structural mismatch between traditional privilege doctrine, which assumes private bilateral communications, and the architecture of many public AI tools that may retain, train on, or disclose user data. The decision signals a new battleground for discovery, potentially shifting privilege risk from content to process and impacting civil litigation, AI copyright cases, and corporate investigations.

Key takeaway

For legal professionals and corporate counsel managing sensitive information, this ruling underscores the critical need to re-evaluate AI tool usage in privileged workflows. You should assume that any information processed through public AI platforms could become discoverable. Implement strict protocols for AI use, favoring enterprise-grade, contractually confidential AI solutions over consumer tools to maintain privilege and avoid inadvertent disclosures that could compromise legal strategy or internal investigations.

Key insights

Using public AI tools for legal prep may waive attorney-client privilege due to third-party disclosure.

Principles

• Privilege requires private, bilateral communication.
• Third-party disclosure typically voids confidentiality.
• Lawyer involvement is key for work product protection.

Method

The government's argument template for challenging privilege involves asserting the AI tool is not a lawyer, the interaction is not for legal advice, confidentiality is lost via third-party platforms, and privilege cannot be retroactively applied.

In practice

• Expect discovery demands for AI prompt logs.
• Assume every AI prompt could be court evidence.
• Prioritize high-control AI environments for legal work.

Topics

• Legal Privilege
• AI in Legal Practice
• Discovery Processes
• Data Confidentiality
• Work Product Doctrine

Best for: VP of Engineering/Data, Director of AI/ML, Executive, Legal Professional, CTO, AI Ethicist

Related on AIssential

• United States v. Heppner — Client-AI communications are not inherently privileged, but a fact-dependent analysis is crucial.
• Simmons & Simmons Launches AI + Legal Privilege Guide — International law firm Simmons & Simmons has launched an "AI and Legal Privilege Guide and Policy Framework" to help manage the risks AI poses to legal privilege, specifically the protection of confid…
• Privilege Saved, Facts Still Exposed: Judge Stein’s OpenAI Dataset Deletion Ruling—and What It Means for the Next Wave of AI Copyright Lawsuits — Denying willfulness or stating non-privileged facts does not waive attorney–client privilege in AI copyright disputes.
• When Conversations with AI Become Evidence — AI chatbot interactions are becoming discoverable legal evidence, challenging privacy and existing evidentiary standards.
• Webinar: AI-Native Workflows in Litigation + Beyond — The core challenge for legal professionals is shifting from AI-enabled tasks to consistent, AI-native workflows.
• Introducing Computer for Counsel - Perplexity — Perplexity's Computer for Counsel automates legal administrative tasks by integrating AI with trusted legal sources and existing workflows.

Open in AIssential →

Editorial summary, takeaway, and curation by AIssential. Original article published by Pascal’s Substack.