OCOPD Internal Evidence Raises Pressure on AI Training Fair-Use Arguments
The legal defense surrounding AI training fair use faces additional scrutiny after internal communications and sworn testimony reported by The Decoder appeared to conflict with some of the technology industry's public arguments about competition and market impact.
A Microsoft director described AI training as the “largest theft of labor in human history,” according to the report. The article also attributed the statement that OpenAI's products “are largely substitutive, period” to the company's head of ChatGPT. The news item identifies the speakers by their Microsoft and OpenAI roles but does not provide their names. It describes the evidence as coming from internal emails and sworn testimony, rather than presenting the statements as public policy positions issued on behalf of either company.
Those remarks are not legal findings. They could nevertheless become useful to copyright plaintiffs arguing that AI developers understood their products as commercially competitive with existing creative markets. The evidentiary significance will depend on the surrounding documents, the questions asked during testimony and whether a court considers the statements relevant to the specific uses and products at issue.
Training Use and Product Substitution Are Separate Questions
Fair-use analysis generally considers the purpose and character of a use, the nature of the copyrighted work, the amount used and the effect on the market for the original. AI companies have argued that training processes extract statistical information from works to create a different technological product, rather than distributing the works themselves.
A statement that an AI product is “substitutive” does not, by itself, establish that a training copy is unlawful. Product substitution and output reproduction are related but distinct questions. A plaintiff may argue that a model competes with authors, publishers or other rights holders even when the model does not normally provide users with a complete copy of a book or article. A company, in turn, may argue that the training use has a different purpose from the market served by the original work.
The reported statements therefore add potential evidence to disputes over commercial purpose and market effects, but they do not resolve the fair-use factors or establish liability across all AI systems.
Bartz and Kadrey Require Case-Specific Treatment
The cases involving Anthropic and Meta should not be treated as interchangeable precedents. Bartz v. Anthropic and Kadrey v. Meta are separate proceedings involving different defendants, records and litigation issues.
Public Knowledge's summary says both decisions treated AI training as potentially transformative fair use and distinguished that question from the acquisition and retention of pirated books. Its account says the courts did not treat the use of pirated source copies as automatically eliminating a fair-use defense for the training use, while leaving potential liability for obtaining and keeping books that could have been obtained lawfully. The summary does not establish that the courts reached identical reasoning, factual findings or outcomes on every issue.
That distinction matters. A ruling about the purpose of training does not necessarily decide whether a company infringed by acquiring source material, nor does it settle claims involving model outputs, market substitution or the removal of copyright-management information. The Public Knowledge analysis is available at “Piracy vs. Fair Use: How AI Training Intersects With Copyright Law”.
Policy Requests Reflect Unsettled Litigation Risk
The industry is also seeking policy changes. Reporting on comments submitted during development of the White House AI Action Plan says OpenAI and Google requested that AI companies receive permission to use copyrighted material for training without paying licensing fees. The narrower point is that the companies sought a policy framework supporting free training use; this should not be described as an established government authorization or as unrestricted permission to copy any copyrighted work.
The policy discussion was reported by the John S. Knight Journalism Fellowships at Stanford, which describes the comments as requests made in response to the White House's AI Action Plan process. The comments and their precise proposed language should be reviewed directly before publication or litigation arguments rely on them.
Business Impact for Model Developers and Rights Holders
The immediate consequence is greater pressure on companies to reconcile internal descriptions, litigation positions and public claims about how AI products compete with creative works. The dispute could affect:
- Training-data costs: Licensing or negotiated access could become more important if courts reject broad fair-use arguments in particular contexts.
- Litigation discovery: Emails and testimony may be examined for evidence about commercial purpose, expected substitution and data-acquisition practices.
- Product controls: Developers may face stronger incentives to reduce memorization and outputs that closely reproduce protected material.
- Policy negotiations: Publishers, authors and other rights holders may push for compensation or transparency rules while companies seek clearer statutory protection.
The evidence reported by The Decoder increases the litigation value of internal language, but it does not supply a universal answer to the copyright question. Courts must still assess the specific training conduct, source material, product markets and evidence in each case.
Source
Original source: the-decoder.com