THE NEW YORK TIMES COMPANY, Plaintiff, v. MICROSOFT CORPORATION, OPENAI, INC., et al., Defendants 23-cv-8292 (SHS) (OTW), 23-cv-11195 (SHS) (OTW) United States District Court, Southern District of New York Filed April 01, 2025 Wang, Ona T., United States Magistrate Judge OPINION & ORDER Pending now before the Court is Defendant OpenAI’s request for a protective order prohibiting the disclosure of certain “highly confidential documents” to Dr. Ricardo Baeza-Yates (“Dr. Baeza-Yates”), an expert retained by Plaintiffs The New York Times and Daily News (collectively, “Plaintiffs”).[1] For the reasons discussed below, OpenAI’s motion is DENIED. I. BACKGROUND The Court assumes familiarity with the facts of this case. On October 18, 2024, OpenAI filed a motion seeking a protective order to prevent the disclosure of confidential documents to Dr. Baeza-Yates, the co-founder and Chief Scientific Officer of Theodora.AI. (24-CV-3285, ECF 163).[2] Plaintiffs filed their opposition on October 22, 2024, arguing that: (1) there is no risk of competitive harm to OpenAI from disclosing documents in this case to Dr. Baeza-Yates; and (2) the Flores balancing test, if applicable, favors rejecting the requested protective order. (ECF 271; 24-CV-3285, 172). On March 3, 2025, Plaintiffs filed an additional letter seeking resolution of the dispute. (ECF 469). The Court notes that that the parties did not include this issue in their last two joint dispute charts filed on January 17, 2025, and February 13, 2025. (See ECF 431, 462). II. ANALYSIS OpenAI argues that a protective order is needed because Theodora.AI is a competitor that “works with revolutionary AI to create deep tech and develop AI-based tools,” and implies that the disclosure of such documents to Theodora.AI’s CSO may cause significant competitive harm to OpenAI. (24-CV-3285, ECF 163, internal citations omitted). Under Fed. R. Civ. P. 26(c)(1)(G), a court may issue a protective order to require “a trade secret or other confidential research, development, or commercial information not be revealed or be revealed only in a specific way.” The party seeking a protective order has the burden of establishing “good cause” for the order, which is established when there is a “particular need for protection” and when a party can show that “disclosure will result in a clearly defined, specific and serious injury.” Ultimately, though, the court has to weigh the interests of both sides. Flores v. Stanford, 18-CV-02468, 2021 WL 4441614, at *4 (S.D.N.Y. Sept. 28, 2021). If the commercially sensitive materials are relevant to a party’s claim, the court must balance the “[receiving] party’s need to have its chosen expert review the materials against the producing party’s interest in protecting the materials from a potential competitor.” Id. at *4.[3] Courts weighing whether to enter a protective order for commercially sensitive materials consider following factors: (1) whether the person receiving the confidential information is involved in competitive decision making or scientific research relating to the subject matter of the materials; (2) the risk of inadvertent disclosure of proprietary information; (3) the hardship imposed by the restriction; (4) the timing of the remedy; and (5) the scope of the remedy. Flores, 2021 WL 4441614, at *4; see also Uniroyal Chem. Co. Inc. v. Syngenta Crop Prot., 224 F.R.D. 53, 57 (D. Conn. 2004). Court must also consider the “specific expertise” of the consultant and whether other consultants “possess similar expertise.” Flores, 2021 WL 4441614, at *4. As an initial matter, OpenAI’s motion should be denied because OpenAI failed to articulate any concrete injury that may result from disclosure of documents to Dr. Baeza-Yates. OpenAI only offers conclusory statements that disclosing documents to Dr. Baeza-Yates will result in commercial harm and that the risk of harm outweighs Plaintiffs’ need to disclose, but OpenAI does not define or quantify what harm is likely to result. (24-CV-3285, ECF 163 at 1). Notwithstanding, the Court will analyze each of the Flores factors in turn. A. Competitive Risk Here, disclosing confidential documents to Dr. Baeza-Yates is unlikely to result in competitive harm to OpenAI simply because Theodora.AI is not a meaningful competitor. Unlike OpenAI, which was valued at $90 billion and generated revenue to the tune of $1 billion just two years ago with innumerous AI-related products, only a portion of which are at issue in the current cases, (e.g., GPT models), Theodora.AI is a small, six-person Chilean company that only works on fine-tuning others’ LLMs to mitigate Spanish language bias. (ECF Nos. 1 ¶ 57, 271 at 1). OpenAI does not allege to have even a single product that is aimed at detecting and fixing biases in Spanish texts. (24-CV-3285, ECF 163).[4] Unlike OpenAI, Theodora.AI does not develop foundational models, pre-train LLMs, or use retrieval augmented generation techniques. (ECF 271 at 1, 3). Theodora.AI also does not fine tune any of OpenAI’s models. (ECF 271-1 ¶ 5). Though Theodora.AI is broadly positioned in the field of artificial intelligence, the company conducts its work in a much narrower scope and, thus, is neither an actual nor potential competitor to OpenAI. OpenAI’s apparent view that because Theodora.AI simply exists in the same space of developing and working on AI-based tools, they must be a competitor with OpenAI, is wildly overbroad. To hold otherwise would essentially make any private company even tangentially related to the AI field, as well as any of their C-suite executives, employees, or vendors, a “competitor” of OpenAI.[5] Even if Theodora.AI were an actual or potential competitor to OpenAI, OpenAI has failed to sufficiently allege that Dr. Baeza-Yates is involved in competitive decision making. OpenAI has provided next to no information as to the extent to which Dr. Baeza-Yates engages in competitive decision making other than the fact that he serves as CSO of Theodora.AI. Defendant cites to no case law indicating that serving as CSO in any capacity creates a presumption of competitive decision making while Plaintiffs indicate that Dr. Baeza-Yates only works in a very limited advisory capacity for approximately 2 hours a week and does not participate in the development work. (ECF 271-1). Additionally, the Defendant has not offered any meaningful reasons to doubt the sufficiency of the existing Protective Order and the confidentiality provisions therein, particularly when the information is being reviewed by a company that does not meaningfully compete with them. Accordingly, the first factor weighs heavily in favor of denying the request for a separate protective order, and courts generally consider this factor as “arguably the determinative factor in this analysis.” Infosint S.A. v. H. Lundbeck A.S., 06CIV2869LAKRL, 2007 WL 1467784, at *3 (S.D.N.Y. May 16, 2007). B. Risk of Inadvertent Disclosure That Theodora.AI is not a competitor to OpenAI greatly informs the Court’s analysis for the second and third factor of the Flores test because involvement in competitive decision making increases the risk of inadvertent disclosure and, if so, needs to be balanced against the hardship the expert’s exclusion would impose.” Infosint S.A, 2007 WL 1467784, at *3; see also Errant Gene Therapeutics, LLC v. Sloan-Kettering Institute for Cancer Research, 15-CV-2044 (AJN) (RLE), 2016 WL 4618972, at *3–4 (S.D.N.Y. Sept. 2, 2016). Analysis of the risk of inadvertent disclosure is typically triggered only if the first factor is met and “when the proposed recipient of sensitive information has an active relationship with one [of] the disclosing party’s competitors.” Flores, 2021 WL 4441614, at *9. In analyzing risk of inadvertent disclosure, courts have found that protective orders are sufficient to protect against such risk and can counter concerns of competitive injury. Flores, 2021 WL 4441614, at *8. Here, because Theodora.AI is not a meaningful competitor with OpenAI, there is little indication that there is a risk of inadvertent disclosure. Even if there were any competition between the two companies, the risk of inadvertent disclosure is minimized given that Dr. Baeza-Yates has signed the existing Protective Order in this case. OpenAI’s “guarantee of advertent disclosure” is conjecture at best, and there is currently no reason to doubt Dr. BaezaYates’s intent to abide by the Protective Order. Though Defendant argues that it would be “impossible” for Dr. Baeza-Yates to separate the disclosed information from his work at Theodora.AI, fear that an expert could accidentally misuse sensitive information because it is in their head, without more, “is too theoretical to warrant preclusion of that expert's receipt of relevant information.” Flores, 2021 WL 4441614 at *10. Accordingly, the second factor also weighs in favor of denying the protective order C. The Hardship Imposed by the Remedy The third factor also favors Plaintiffs because granting a protective order will deny them the benefit of having their chosen consultant review the materials with his particular background and expertise. Plaintiffs will be harmed if Dr. Baeza-Yates, as their chosen consultant, is unable to review the material. (ECF 271 at 3). Defendant argues that Vinay Hooloomann and Professor Oren Etzioni can serve as alternatives to Dr. Baeza-Yates. Their areas of expertise, however, do not entirely overlap with Dr. Baeza-Yates. Mr. Hooloomann is a source code reviewer who does work on the development of LLMs using retrieval-augmented generation (RAG), but his CV indicates that he began such work in April 2023 and works primarily in the context of healthcare. (24-CV-3285, ECF 166 at 194). Dr. Baeza-Yates, on the other hand, has published extensively on web search, web mining, and information retrieval and worked at Yahoo Labs for 10 years doing work on web search, usage data mining, web advertising. (24-CV-3285, ECF 166 at 9–44). Professor Etzioni is primarily involved in research and academia. (24-CV-3285, ECF 166-4, ECF 166-5).[6] Dr. Baeza-Yates’s experience on information retrieval outside of academia could provide unique insight and make him “particularly well-suited to address issues related to retrieval-augmented generation, among other issues. . .” (ECF 271 at 3). Accordingly, the third factor also weighs in favor of denying the protective order. D. Timing of the Remedy The fourth factor favors, albeit slightly, the Defendant. If the remedy of a protective order is granted, it would not substantially impact the efficiency of discovery thus far. If history is any indication of the future of discovery in this case, any delay from the granting of a protective order will be minimal in scope. E. The Scope of the Remedy The final factor does not favor either party. The remedy that OpenAI seeks is narrow in that it would only block disclosure to one of Plaintiffs’ disclosed experts. However, as discussed above, the requested remedy is also manifestly overbroad because it would effectively render any private company that works in AI a “competitor,” and thus block additional experts that Plaintiffs may seek to replace Dr. Baeza-Yates with from accessing material and potentially testifying. Accordingly, the final factor does not weigh for or against granting a protective order. F. Weighing the Factors After considering the five Flores factors and having given them the appropriate weight, I find that OpenAI’s request for a protective order as to documents produced to Dr. Baeza-Yates is not warranted. Accordingly, Defendant’s motion is DENIED. III. CONCLUSION For the reasons stated above, OpenAI’s motion for a protective order to prevent disclosure of confidential information to Dr. Baeza-Yates is DENIED. The Clerk of Court is respectfully directed to close Case No. 24-CV-3285, ECF 163. SO ORDERED. Footnotes [1] While the New York Times and Daily News cases were consolidated at the September 12, 2024, status conference, (see ECF 243), OpenAI’s motion for a protective order was filed only on the Daily News docket. (24-CV3285, ECF 163). Plaintiffs filed a responsive letter on both dockets. (ECF 271; 24-CV-3285, ECF 172). Plaintiffs then filed a letter on the New York Times docket informing the Court that the parties met and conferred by remained at an impasse. (ECF 469). [2] Though the motion refers to Dr. Baeza-Yates’s as Daily News’s expert, The New York Times subsequently disclosed Dr. Baeza-Yates as an expert as well. [3] The existing Protective Order in these consolidated cases also sets forth: [When a party seeks] to disclose to a testifying or consulting expert any information or item that has been designated ‘HIGHLY CONFIDENTIAL – ATTORNEYS’ EYES ONLY’ or ‘HIGHLY CONFIDENTIAL – SOURCE CODE’ [then the] party opposing disclosure to the expert shall bear the burden of proving that the risk of harm that the disclosure would entail (under the safeguards proposed) outweighs the recipient’s need to disclose the Protected Discovery Material to its expert. (ECF 127 ¶ 17(b); 24-cv-3285, ECF 129 ¶ 16(b)). The parties’ original protective order in this case was filed on May 31, 2024. (ECF 127). The parties recently submitted a revised protective order, which the Court entered on March 18, 2025. (ECF Nos. 473, 474). The revised protective order does not affect the analysis in this order. [4] OpenAI’s has also stated that the “majority of [OpenAI’s] pretraining data and… alignment data is in English” and “mitigations and measurements [of GPT-4] were mostly designed, built, and tested primarily in English and with a US-centric point of view.” (ECF 271 at 1). [5] See Advanced Micro Devices, Inc. v. LG Elecs., Inc., 14-CV-01012, 2017 WL 3021018, at *3 (N.D. Cal. July 17, 2017) (“However, neither [expert] works directly for an Imagination competitor, and Imagination has provided no concrete evidence that either one engages in or assists with competitive decision-making at an Imagination competitor. The Court finds it unlikely that AMD would be able to select an expert in this space who does not consult, publish, teach, or invent in the relevant field.”). [6] Professor Melanie Mitchell, whose expertise is disclosed in the Nightingale Declaration, was not proffered as an alternative by OpenAI.