In James v. Cerebras Systems Inc., the Northern District of California endorsed closed AI environments for sensitive discovery materials by approving a stipulated protective order that directly addresses the use of large language models (“LLMs”) and generative AI tools in litigation. The order reflects growing judicial concern over the risks associated with uploading confidential discovery materials into AI platforms and establishes guardrails for litigants seeking to leverage AI-assisted workflows. Most notably, the court required parties to take commercially reasonable steps to prevent unauthorized access to protected material when using generative AI tools and prohibited the use of protected information in AI systems unless users ensure that the information will not be incorporated into model training and that the platform maintains industry-standard cybersecurity protections. The order specifically recognized enterprise versions of ChatGPT and Harvey as satisfying those requirements, signaling judicial preference for closed, controlled AI environments over public-facing generative AI platforms.

Although entered as a stipulated protective order rather than a contested discovery ruling, James is significant because it provides one of the clearest judicial endorsements to date of “closed” AI systems for handling sensitive litigation data. The order signals that legal teams can no longer treat generative AI as a generic productivity tool. Instead, counsel must understand where data is processed, whether prompts and documents are used to train models, who can access the information, and what security controls are in place. These requirements will affect every stage of the eDiscovery lifecycle, including document review, privilege analysis, deposition preparation, case strategy development, and AI-assisted summarization. Organizations relying on public-facing AI tools will face increased scrutiny regarding confidentiality obligations, protective order compliance, and cybersecurity safeguards. As courts continue to address the risks posed by generative AI, James demonstrates that defensible legal workflows will increasingly require enterprise-grade, closed AI environments with robust governance controls and clear auditability. Parties seeking to leverage AI while minimizing risk should ensure their workflows align with these evolving expectations.

If you would like assistance implementing an AI-enabled eDiscovery workflow that complies with the increasingly strict judicial requirements surrounding closed AI systems and protected data, reach out to Nicholas Berenato at CODISCOVR. At CODISCOVR, we deliver client-focused, defensible solutions that are tailored to each organization’s specific needs. Nick is an attorney who collaborates with clients to address legal and financial exposure from their information assets through information governance. He develops strategies and policies to enable clients to effectively manage their data, minimize costs, and leverage it to support their business processes.