Short on time? Click here to listen to this article on the Sovereign Discovery website. |
S|D | SOVEREIGN DISCOVERY eDiscovery & Litigation Support |
|
Newsletter · Issue 3 Your AI Prompts Are Now Discoverable. Is Your Legal Hold Ready? Three federal courts in 90 days. One message: AI interactions are ESI, and most legal hold notices don't say a word about them. |
Something significant happened in the first quarter of 2026 that most legal teams have not yet acted on. Three federal courts — in New York, Michigan, and Colorado — issued rulings that, taken together, establish a new reality: the interactions your employees have with AI tools are electronically stored information. They are subject to preservation obligations, potentially discoverable, and in some cases already stripping away privilege protections attorneys assumed were safely in place. The gap between how organizations are using AI and how they are managing the discovery implications of that use is growing — and courts are beginning to notice. |
S.D.N.Y. · Feb. 17 US v. Heppner AI materials created without attorney direction do not qualify for privilege or work product protection. | | E.D. Mich. · Feb. 10 Warner v. Gilbarco Disclosing to an AI tool does not constitute disclosure to an adversary — work product waiver denied. | | D. Colo. · Mar. 30 Morgan v. V2X Cannot upload confidential discovery data to AI tools unless the provider is barred from training on it. |
|
Visual Overview Three Courts. Three Rulings. One Gap in Every Legal Hold. LITIGATION HOLD NOTICE — CURRENT STATE EMAIL & CALENDAR COVERED | | SHARED DRIVES COVERED |
SLACK & TEAMS COVERED | | AI INTERACTIONS & OUTPUTS Not addressed in this hold |
This gap is now a documented court concern — Heppner · Warner · Morgan ChatGPT prompts · Copilot sessions · Gemini outputs · AI-assisted drafts |
|
01 | What Happened Three Courts Built a Framework in 90 Days |
|
In Heppner, the Southern District of New York ruled that AI-generated materials created without attorney direction do not qualify for attorney-client privilege or work product protection. The court found that gap dispositive. In Warner, the Eastern District of Michigan reached a different result. A pro se plaintiff had used generative AI to prepare litigation materials. The court denied the motion to compel, reasoning that AI platforms are tools — disclosing information to them does not constitute disclosure to an adversary sufficient to waive work product protection. Then in March, Morgan v. V2X issued the most comprehensive ruling of the three. The court established that parties cannot upload confidential discovery materials to consumer AI tools unless the provider is contractually barred from using that data for model training. |
After Heppner, Warner, and Morgan, AI prompts and outputs clearly fall within the scope of potentially relevant ESI that must be preserved once litigation is reasonably anticipated. |
|
02 | Why It Matters The Practical Implications Are Immediate |
|
Most litigation hold notices still do not address AI interactions. After Heppner and Morgan, that gap is difficult to defend. Firms need to audit whether AI tools retain user input, whether that input constitutes ESI, and whether current hold protocols capture it. Corporate legal departments face a layered risk. Employees across organizations use AI tools every day — for drafting, summarizing, analyzing, and communicating. If litigation is reasonably anticipated, existing hold notices likely fail to account for AI-generated content. The Morgan decision also functions as a data security warning. Uploading confidential business data into consumer AI tools that may use it for model training is now a court-identified risk. Protective order language will need to address AI tool usage going forward. |
Section 2 · Visual What Your Legal Hold Needs to Say Now 1 | Identify which AI platforms employees are using ChatGPT, Copilot, Gemini, Claude, Notion AI, and any embedded AI features in business tools |
2 | Determine which tools retain user input and for how long Not all AI tools store interaction history — but many do. This determines your hold scope. |
3 | Add AI-specific language to all litigation hold notices Cover prompts, outputs, AI-assisted drafts, and any documents where AI contributed to content |
✓ | Audit vendor agreements for AI data training restrictions Per Morgan — if confidential data may enter AI tools, the provider must be contractually barred from training on it |
|
|
Key Takeaways AI prompts and outputs are ESI. They must be preserved once litigation is reasonably anticipated. Privilege depends on attorney direction. Undirected AI use is now legally significant and at risk. Consumer AI + confidential discovery data don't mix. Morgan makes this a protective order issue. Legal hold notices must be updated now. The gap is documented — and courts are watching. |
|
Industry Perspective — Sovereign Discovery What Heppner, Warner, and Morgan confirm is that the conversation legal teams have been deferring — about how AI use fits into their discovery and privilege frameworks — is no longer optional. The organizations that will navigate this well are those that treat AI governance as a discovery readiness issue — not just a technology policy. That means understanding data flows, reviewing hold language, auditing vendor agreements, and ensuring that anyone touching AI tools in a litigation context understands the implications. |
|
Final Thought The courts are not waiting for the legal industry to develop consensus on AI and discovery. They are issuing decisions now — case by case, jurisdiction by jurisdiction. The question is no longer whether your organization uses AI. The question is whether your discovery process accounts for it. |
Is your legal hold ready for AI-generated ESI? Schedule a Consultation |
Sovereign Discovery | 1101 15th St, NW · Washington, DC 20005
eDiscovery & Litigation Support |
|
| You received this because you subscribed to Sovereign Discovery updates. · Unsubscribe |