In a previous installment, we navigated the emerging storm of legal holds and generative AI, outlining the new preservation duties for in-house counsel. We promised a practical playbook. But first, a necessary descent into Dante's Inferno, AI Edition.
Seven Hellscapes You Can't Ignore
The duty to preserve electronically stored information (ESI) has been dramatically expanded by AI. Here are the critical risks that have emerged:
- The Vanishing Act Problem - Unlike a traditional email that lives on a server, AI outputs are often ephemeral. That brilliant, AI-generated marketing campaign? It can vanish with a single "regenerate" command, making it as permanent as a Snapchat message.
- The "Whose Cloud Is It Anyway?" Dilemma - Your data is no longer confined to your servers. It's now scattered across a fragmented ecosystem of third-party vendors like OpenAI and Anthropic. Explaining to opposing counsel why you can't produce data that isn't technically in your custody will be an uphill battle.
- The Privilege Panic - A senior attorney who pastes confidential strategy into ChatGPT to "polish the language" might have just waived attorney-client privilege. Without careful controls, AI tools do not provide built-in privilege or work product protection.
- The Authentication Nightmare - "Your Honor, we'd like to introduce this AI-generated analysis into evidence." The inevitable follow-up question: "Can you prove who created it, when, and that it hasn't been altered?" Proving the provenance and integrity of AI outputs is a significant and often difficult task. The emerging need for metadata preservation standards for AI outputs has not yet caught up to legal requirements.
- The Volume Volcano - You thought managing thousands of emails was a challenge? Now, imagine adding every AI interaction across your entire organization—from prompts to chat histories to outputs. Your document review costs and timeline may have just grown (again).
- The Training Data Time Bomb - Many AI platforms use your prompts to train their models. That confidential merger strategy could be teaching AI systems worldwide. Deleting the conversation doesn't undo the fact that your data may have already been used.
- The "We Don't Log That" Response - When opposing counsel asks for all ChatGPT conversations from March 2024, and the provider says "we don't keep those," your company risks looking like it's hiding something. The burden of preservation (usually) falls on you, not the vendor.
AI Requires Governance
To survive an AI-related legal hold, your company needs a phased action plan. Start with immediate "quick fixes," then move to more sustainable policies and systems.
Today's "Oh Crap" Fixes: Immediately send a revised legal hold notice that specifically includes AI tools, such as prompts and outputs. Follow up with an all-hands email instructing employees to stop deleting any AI conversations. Launch a 48-hour sprint to inventory all AI tools currently in use across the organization.
This Week's Actionable Steps: Draft and implement an AI usage policy that mandates business accounts, requires prompt documentation for sensitive matters, and bans the use of privileged and other sensitive information. Collaborate with your IT team to identify and test preservation workflows for each tool. Review vendor agreements to understand data retention and export capabilities.
This Month's Sustainable Solutions: Upgrade your legal hold software to one that can handle AI-specific preservation. Develop a protocol for authenticating AI evidence, including chain-of-custody and screenshot standards. Finally, train all employees on what constitutes AI-generated ESI and their preservation obligations to ensure company-wide compliance.
The Bottom Line (Because You're Busy)
AI isn't going away, and neither are your legal hold obligations. The courts are catching up, opposing counsel is getting savvier, and the "we didn't know" defense is no longer valid (if it ever was).
The good news? You're not alone. The bad news? That won't save you from sanctions if you get it wrong. Start small, be systematic, and remember: in the age of AI, paranoia isn't a bug—it's a feature.
P.S. - Yes, this blog post was ideated, outlined, and brooded over with the assistance of AI. Yes, we preserved the prompts. Yes, we're practicing what we preach. No, we're not perfect at it yet either.
The content of this article is intended to provide a general guide to the subject matter. Specialist advice should be sought about your specific circumstances.
