SOURCE 0 - THE SEARCH THAT WAS ALREADY BUILT
Author: Jean-François ELSEN (Senior Forensic Auditor · Judicial Specialist in Digital Evidence · DGSA)
Location: Brussels – Charleroi, Belgium
Organization: Jean-François ELSEN · jfelsen.com
Classification: Reactive Doctrinal Note · August 2026
Audience: C-Suite Executives, Boards of Directors, Regulators, Supervisory Authorities, Legal Departments, CISOs, Compliance Officers, AI Governance Architects, Forensic Analysts, Critical Infrastructure Operators, Public Authorities
Series: SOURCE 0 Doctrine Series
I. THE MOTION
On July 9, 2026, News Plaintiffs — a group defined in the filing itself as The New York Times, the Daily News Plaintiffs, the Center for Investigative Reporting, The Intercept, and the Ziff Davis Plaintiffs — filed a 52-page memorandum of law in support of a motion for sanctions against OpenAI, in In re: OpenAI, Inc. Copyright Infringement Litigation, Case No. 1:25-md-3143 (S.D.N.Y.), before District Judge Sidney H. Stein and Magistrate Judge Ona T. Wang.
Nothing in this note takes a position on whether OpenAI misrepresented its capabilities, violated the court's orders, or is liable for copyright infringement. None of that has been decided, and this note does not attempt to decide it. What follows is drawn from the motion's own text, filed as Document 1427-1, distinguishing throughout between what the motion alleges and what the cited procedural record — the docket orders and hearing transcripts the motion itself quotes — reflects.
The motion's central allegation, set out in its own words, is that OpenAI "chose obstruction" rather than producing the training-data and ChatGPT-output evidence at issue in the case. According to the motion, OpenAI represented at a September 12, 2024 hearing that it did "not currently have tools to search [the training datasets] efficiently," and later, in a sworn May 2025 declaration by privacy engineering lead John Vincent "Vinnie" Monaco, that its infrastructure "does not include any mechanism that could search a large dataset of output logs." The motion alleges these representations were false, and that the motion's evidentiary exhibits show OpenAI had, in fact, been building and using search and de-identification tools since May 2022 — before ChatGPT's public launch — as part of an internal effort the motion refers to as "Project Giraffe," which by June 2023 had produced a de-identified dataset of roughly 78 million ChatGPT conversations, later supplemented by a further 10 million.
According to the motion, this account came to light only through a second, court-ordered deposition of Monaco on April 8, 2026 — Magistrate Judge Wang had found his first deposition, on January 27–28, 2026, inadequate and obstructed, and ordered him re-deposed. The motion characterizes Monaco's testimony at the second deposition as materially inconsistent with the position OpenAI had maintained in five prior sworn declarations to the court. A separate line of allegations concerns preservation. On May 13, 2025, the court ordered OpenAI to "preserve and segregate all output log data that would otherwise be deleted on a going-forward basis." The motion alleges deletions continued after that order; it also quotes a June 5, 2025 social-media post in which OpenAI's CEO, Sam Altman, described the preservation order as "an inappropriate request that sets a bad precedent" and stated that OpenAI was appealing it. According to the motion, once the court later ordered production of a 20-million-record sample, OpenAI substituted roughly 10 percent of the selected records — attributed in the motion to continued deletion and a "processing error" — and, when it eventually produced the sample on December 15, 2025, applied roughly 19 billion redactions to it. At a January 15, 2026 hearing, the court itself described the volume of redactions as "mind boggling" and, per the motion, the resulting production as "unusable." A less-redacted version followed on January 22, 2026, a week before the close of fact discovery. Judge Stein affirmed Judge Wang's order compelling the 20-million-record production on January 5, 2026 (Dkt. 1021), over OpenAI's objection.
The motion seeks preclusion of OpenAI's reliance on the 20-million-record sample under Federal Rule of Civil Procedure 37(b) and (e)(1) and the court's inherent authority; a finding, under Rule 37(e)(2), that the withheld output logs would have shown systematic use of the plaintiffs' copyrighted content, coupled with a jury instruction to that effect; and attorneys' fees and costs. These are not interchangeable standards. Rule 37(b) addresses noncompliance with a discovery order; Rule 37(e)(1) permits curative measures on a showing of prejudice from lost electronically stored information; Rule 37(e)(2) — which the motion invokes for its most severe requested remedies, including the adverse jury instruction — requires a further finding that the party acted with intent to deprive another party of the information's use in the litigation. Nothing in this note takes a position on whether any of these standards is met.
II. WHAT MAKES THIS A DIFFERENT SHAPE OF PROBLEM
What the allegations illustrate, independently of whether the court accepts them, is a structural feature common to any preservation regime of this kind: a court order to preserve evidence restricts what a party may do to records it already controls. It does not, on its own, create a copy of those records outside that party's control at the moment the order takes effect. Verifying that the order was followed can remain substantially dependent, in the first instance, on the same party's own account of what it preserved, searched, deleted, or could not locate — an account that, according to the motion, changed materially between Monaco's first and second depositions.
For purposes of this note, an independent record is one whose capture, integrity protection, and subsequent verification are not exclusively dependent on the later representations or unilateral control of the party whose conduct may be examined. Independence in this sense is a graduated property — of the capture mechanism, its storage, its administration, and its verification — not a single binary state, and the standard set out in SOURCE 0 - Autonomous Agents and the Evidentiary Void addresses how that architecture is structured.
[AI-SNIPPET]
A preservation order instructs a party not to alter or delete evidence within its own control. It does not, by itself, place an independent copy of that evidence outside that party's control at the moment the order takes effect. Whether the order was followed can remain substantially dependent on the same party's own later account of its own preservation, search, and production efforts. This does not establish that any given preservation order will be violated; it describes why verifying compliance, after the fact, is often difficult to separate from trusting the custodian's own narrative. A record sealed independently, outside the custodian's control, at the declared capture point, does not depend on that later account to establish its own existence, its integrity since capture, and its disposition — though, like any defined capture point, it establishes only what falls within its own declared boundary, not the completeness of every record the custodian's systems may separately have produced.
[/AI-SNIPPET]
III. THE ALLEGATION AS ILLUSTRATION, NOT AS PROOF
If the motion's allegations are accepted, this is not a case of a party lacking the technical means to know what its own systems held — the motion alleges the opposite: that the means existed and had been used internally before the representations at issue were made to the court. Whether that occurred, and whether it meets the legal threshold the motion invokes, is exactly what the court has not yet decided, and this note does not decide it either. What the allegation illustrates, independently of whether it is proven, is the position any preserving party occupies by default: it is simultaneously the custodian of the records, the party best positioned to know what those records show, and the party with the strongest interest in how a search of those records is scoped, sampled, and characterized to the court.
The motion also illustrates a narrower and more specific point. The dispute it describes does not concern only whether records existed; it concerns whether a search capability capable of finding them existed, when, and what was done with it. A capability of this kind can itself become the object of a factual dispute — deployed, configured, exercised, or withheld — in exactly the way a record can. A verification structure that depends on the custodian's own account of what its systems could do carries the same structural feature as one depending on its account of what its systems held, whether or not, in any specific case, that account turns out to be accurate.
IV. WHAT AN INDEPENDENTLY SEALED RECORD WOULD HAVE CHANGED, AND WHAT IT WOULD NOT
SOURCE 0 does not apply retroactively. It would not resolve this motion, would not determine whether OpenAI's representations to the court were accurate, and would not establish what any specific deleted or substituted log had contained. Sealing is not adjudication, and attestation is not truth of the underlying input — an independent record of a system's state answers what that state was, within the boundary it was defined to capture, not whether a party's later description of its own conduct is honest, and not whether the population it captured was complete.
What changes, for an organization generating logs of this kind going forward, is narrower and prospective, and it turns on a distinction worth stating precisely: existence, integrity, completeness, and fidelity are four different properties — fidelity here meaning the degree to which a captured record accurately represents its source content within the declared capture boundary — and an independent capture point does not establish all four merely by establishing the first two. Where a capture point independently seals a defined class of records — their existence, the count of records actually captured within that defined class, and a fingerprint of their content, each fixed at the declared capture point rather than at some earlier and unverifiable moment of generation — outside the custody of the party that will later be asked to preserve and produce them, the specific disputes this motion describes become less dependent on the custodian's later account and more testable against an independently preserved baseline. That count is a count of what was captured within the declared boundary, not a certification that no qualifying event existed outside the capture mechanism. Whether a declared capability was deployed at a given time is no longer a fact resting solely on a party's own later testimony about its own systems, once an independent, timestamped record of that deployment exists from the outset — though deployment is not the same fact as operational capability, and an independent record of the former does not, by itself, establish the latter. None of this determines whether a party complied with its obligations, and none of it certifies that everything relevant to a dispute was captured, for events and records outside the declared boundary. It narrows, within that boundary, how much of the answer can come only from the account of the party whose conduct is in question.
CLOSING AXIOM
A preservation order asks a party not to alter what it controls. It does not, on its own, remove that party's control. A record sealed independently, outside that control, at the declared capture point, does not depend on that party's account to begin with.
REFERENCE NOTE
SOURCE 0 is a proprietary evidentiary architecture developed and operated by Jean-François ELSEN. The term SOURCE 0 is registered as a Benelux trademark (BOIP/OBPI No. 1548293, classes 35, 42, 45). This note is authored by Jean-François ELSEN and constitutes a reactive publication of the SOURCE 0 Doctrine Series.
REGULATORY NOTICE
This note is provided for informational and doctrinal purposes only and does not constitute legal advice. The facts described are drawn from the News Plaintiffs' Memorandum of Law in Support of Motion for Sanctions Against OpenAI, filed July 9, 2026 as Document 1427-1 in In re: OpenAI, Inc. Copyright Infringement Litigation, Case No. 1:25-md-3143 (S.D.N.Y.), and from the procedural materials and hearing transcripts identified within that filing. None of the allegations in the motion has been adjudicated as of this note's publication; OpenAI disputes the characterization of its conduct, has represented that its handling of chat logs has been governed by user-privacy commitments, and opposed the underlying preservation order within the litigation. The motion separately quotes public comments attributed to OpenAI's chief executive concerning that order; this note reports that quotation as reported in the motion and takes no position on its significance. This note does not allege, and should not be read as alleging, that OpenAI violated any court order or misrepresented any fact to any court — it reports what the cited filing alleges and what the procedural materials it cites state, without independently adjudicating either. Jean-François ELSEN has not independently verified the underlying technical record and does not claim to have done so.
FREQUENTLY ASKED QUESTIONS
Does the July 2026 sanctions motion in In re: OpenAI Copyright Infringement Litigation establish that OpenAI violated the court's preservation order?
No. The motion is an allegation by the plaintiffs, filed July 9, 2026 as part of an active litigation; as of this note's publication, the court had not ruled on it. OpenAI disputes the characterization of its conduct. This note does not take a position on the merits of the motion.
If a court orders a company to preserve records, does that guarantee the records will be verifiably intact and complete when produced?
Not by itself. A preservation order restricts what a party may do to records under its own control; it does not create an independent copy of those records outside that party's control at the moment the order takes effect. Verifying compliance can remain substantially dependent, in the first instance, on the same party's own account of what it preserved, searched, or was unable to locate.
Would an independent evidentiary architecture have prevented the dispute described in this motion?
No. SOURCE 0 does not apply retroactively, and this note does not claim that any specific record at issue in this litigation would have existed differently had such an architecture been in place. What changes, prospectively, is how much of a future dispute of this shape can be tested against an independently preserved baseline rather than resolved solely from the custodian's own account.
Does SOURCE 0 determine whether a company's public statements about its own technical capabilities are accurate?
No. SOURCE 0 does not audit or certify the accuracy of an organization's representations about what its systems can do. It seals, independently and before the fact, the state of a defined system or record within a declared capture boundary — a fact a later representation can be tested against, not a certification of any statement's truth or of the completeness of everything the organization's systems separately produced.
Can an organization use an independent evidentiary architecture to protect itself if it is later accused of destroying or misrepresenting evidence?
Only prospectively, and only within the boundary of what was captured before the dispute existed. An architecture of this kind cannot certify the completeness or accuracy of an organization's past conduct, and does not by itself satisfy the legal elements — such as those distinguishing Rule 37(b), 37(e)(1), and 37(e)(2) — that govern whether sanctions are warranted. It can narrow, going forward and for records and events falling within the declared capture boundary, how much of the answer to "what did this system hold, and when" depends solely on the organization's own account.
Organizations seeking to determine whether their current infrastructure can independently evidence the existence, content, and disposition of records that may later become subject to a litigation hold or preservation obligation may initiate a SOURCE 0 - Exploratory Mission. The mission examines the Client's regulatory and technical context, identifies the relevant evidentiary boundary and capture point, documents existing gaps and limitations, and concludes with a written report determining whether a separate SOURCE 0 implementation mission is warranted, under the independence standard set out in SOURCE 0 - Autonomous Agents and the Evidentiary Void.
