TL;DR:
In Chicken Soup for the Soul, LLC v. Meta Platforms, Inc., a Northern District of California court on September 10, 2026 entered a discovery order addressing OpenAI and Anthropic subpoenas, restricting some lines of inquiry into AI training data. The order focuses on three topics—market substitution evidence, licensing evidence, and torrenting evidence—and largely denies several discovery requests, finding the proposals burdensome and not proportional to the needs of the case. The ruling signals tighter controls on how AI training data and related materials can be obtained in high-stakes copyright disputes, with immediate implications for case strategy, privilege, and evidentiary foundations in tech-driven litigation. The order also demonstrates a growing trend toward more granular eDiscovery protocols in AI-related disputes, including considerations of technology-assisted review, data provenance, and the limits of licensing markets for training data.(cases.justia.com)
What happened
The decision arises in a consolidated set of proceedings involving Chicken Soup for the Soul, Meta Platforms, and related actions, with motions to compel directed at OpenAI and Anthropic. The court’s September 10, 2026 Discovery Order, signed by Judge Thomas S. Hixson, addresses requests tied to three discrete topics:
Market Substitution Evidence: Plaintiffs sought documents and testimony about the ability of AI models (including OpenAI’s ChatGPT and Anthropic’s Claude) to generate literary content and how those capabilities might affect markets for licensing or use of human-authored works. The court’s analysis emphasizes questions of relevance and proportionality, noting that complex, broad discovery into model-generated output versus market substitution risks being overly burdensome and not easily tethered to the fair-use questions at issue in Meta’s defense. The court denied the portion of the motion to compel on these market-substitution requests against OpenAI and Anthropic for the reasons stated in the order. (cases.justia.com)
Licensing Evidence: Plaintiffs also sought documents concerning licensing efforts to use books or other content for AI training. The court rejected this line of inquiry against OpenAI and Anthropic, again finding the requests too broad and not sufficiently connected to the core issues in Meta’s fair-use defense in this skeletal stage of discovery. The order explicitly denies this portion of the motion to compel as to both OpenAI and Anthropic. (cases.justia.com)
Torrenting Evidence: Plaintiffs urged production related to torrenting data and related technicalities of data acquisition for AI training. The court denied those requests as to both OpenAI and Anthropic, indicating that the burdens outweighed the likely probative value given the case posture and the need to focus discovery on the most relevant issues at hand. (cases.justia.com)
The order is notable for its careful calibration of AI discovery in a high-profile copyright fight. It shows the court constraining extensive, technically intensive discovery into training data and model development when the connection to the underlying copyright issues is not sufficiently proven or proportionate. The specific form of the ruling reflected a detailed, page-by-page analysis of each contested topic, rather than a broad open-ended fishing expedition. The order is publicly accessible via court filings and the docket records, including a PDF version of the decision. (cases.justia.com)
Why this matters for trial attorneys
Narrowing the scope of AI discovery in a major copyright dispute: The order demonstrates that courts are willing to curb expansive demands for training-data materials, model internals, and related licensing discussions when the direct relevance to the anticipated trial issues is not adequately demonstrated. For trial teams, this means less risk of overbroad production requests ballooning into a sprawling, expensive discovery phase with technical experts and data scientists driving the process. (cases.justia.com)
Implications for evidentiary foundations around AI-generated content: Because the litigants sought to establish market-substitution and licensing dynamics as part of a fair-use analysis, the court’s cautious stance signals that similar evidence in other cases may face heightened scrutiny for relevance and proportionality. Practitioners should frame such AI-related evidence around clearly traceable issues, for example how a particular output might be used in court, rather than broad assertions about market impact of AI training. (cases.justia.com)
Privilege and work product considerations become central in AI disclosure: The court’s approach underscores the need to carefully segregate privileged communications and work product from non-privileged AI-discovery materials, especially where technical analyses and model assessments could reveal drafts, training data discussions, or evaluation methods. Anticipate robust privilege logs and targeted clawbacks where appropriate. (cases.justia.com)
Practical workflow implications for eDiscovery teams: With the court emphasizing proportionality and limiting burdens, litigators should tailor their discovery plan to focus on the most probative AI-related data points. This includes coordinating with technical experts to identify the narrow set of materials most likely to influence the court’s fair-use or copyright determinations, and to prepare clean, defensible affidavits or declarations that demonstrate relevance. (cases.justia.com)
Immediate steps for trial teams handling AI-driven evidence: Map out your core issues in the case, identify the precise AI-related items you would need to prove or rebut a claim, and build a focused discovery matrix. Prepare to discuss with opposing counsel and the court any proposed limitations or protections to address sensitive training-data information, and ensure that your objection and foundation strategies are ready for a potential evidentiary hearing. Consider leveraging practical AI-evidence training resources to sharpen these skills in real-world settings. (cases.justia.com)
Practical next steps for litigators
Tighten the line of inquiry: If pursuing AI training-data discovery, align requests with a specific theory of relevance tied to the case’s central issues, such as a particular AI output’s potential relevance to a copyrighted work or licensing landscape, rather than broad, generalized inquiries into model capabilities.
Prepare precise limits and protocols: Anticipate requests for procedural or technical safeguards, including potential sampling plans, privilege protections, and data-use limitations. The court’s decision illustrates the value of proposing transparent, well-structured discovery protocols up front.
Plan for expert testimony and cross-examination: In AI disputes, expert witnesses on machine-learning methodology, data provenance, or licensing markets may be essential. However, be ready to defend the necessity and limits of such testimony in light of proportionality concerns highlighted by the court.
Leverage training resources for AI-evidence readiness: Trial teams may benefit from targeted training on AI-generated content, chain-of-custody issues, and objection strategies. Training platforms focused on AI-evidence foundations and objection handling can help teams prepare for the evolving evidentiary landscape.
Consider parallel ethics and compliance work: Given the cross-cutting concerns around AI, training data, and discovery, consider aligning case strategy with internal ethics and client compliance considerations. Courts are increasingly addressing how counsel should handle AI-generated materials, and proactive preparation can smooth the litigation path.
For ongoing AI-evidence readiness, litigation teams can complement their strategy with specialized practice resources that focus on AI-generated content, objections, and trial-readiness. While the legal debate about AI training data continues to evolve, disciplined discovery planning and rigorous evidentiary foundations remain essential to presenting AI-related issues effectively at trial.
Sources:
- Chicken Soup for the Soul, LLC v. Meta Platforms, Inc., Discovery Order (Sept. 10, 2026). PDF and docket details, including the open and contested topics (OpenAI and Anthropic subpoenas). See Case No. 26-cv-02333, 26-cv-03725, 26-cv-04053 (Northern District of California). (cases.justia.com)
- Chicken Soup for the Soul, LLC v. Meta Platforms, Inc., Discovery Order (Sept. 10, 2026) – full text with discussion of market substitution, licensing, and torrenting evidence. (cases.justia.com)
Note: This analysis reflects the September 2026 order and its immediate implications for AI-related discovery in high-stakes copyright litigation. For practitioners, it underscores the importance of precise, proportional, and well-documented discovery practices when AI training data and model outputs become central to the case.