AI-generated content is produced with a single click, but it does not come out of thin air. Behind every AI-generated response is work that humans created first. Yet the output carries no trace of its source: no name, no payment, and no way for creators to prove the work was theirs. Instead, the “AI-generated” label makes the people behind the idea disappear.
Take Doe v. GitHub. Programmers argued that GitHub’s Copilot reproduced their code without their names or any information about them or their work. On September 16, 2026, the Ninth Circuit disagreed, treating Copilot’s output as new code rather than a copy stripped of its credits. The machine ends up at the center, and the humans remain invisible. Transparency rules should not stop at disclosing AI use. They should also identify the human work behind the output, so credit goes to the people who deserve it.
Governments are now deciding who owns the material AI learns from, and that question shapes who owns what it produces. The Justice Department intervened in the OpenAI copyright lawsuit in New York on September 1, 2026, arguing that using copyrighted material to train AI constitutes fair use. According to Axios, the Patent and Trademark Office and the Copyright Office were surprised by the Justice Department’s filing, and the government’s fair use argument applied to AI training rather than AI output. The Third Circuit ruled in Thomson Reuters v. Ross that reproducing copyrighted works to build a competing product did not qualify as fair use on September 29, 2026. Because that case involved non-generative AI, though, it left open how fair use applies to models like ChatGPT, and creators seeking compensation still don’t have an answer.
Broader questions about generative AI may ultimately require legislative resolution. Europe has gone the opposite way; that’s because the European Parliament on March 10, 2026, adopted a resolution demanding mandatory disclosure, requiring AI providers to reveal which copyrighted works were used in training data sets, with a rebuttable presumption that undisclosed works were used if the provider cannot adequately document its training data. If enacted, the framework would cover AI models available in the European Union, regardless of where they were trained; however, it remains a proposal, not a law. The UK likewise stopped short of adopting an opt-out system; rather than facing objections from creators and publishers, the March 18, 2026 report dropped plans for the opt-out system and focused on gathering evidence and developing voluntary transparency standards.
Beneath the legal debate over AI training data lies a quieter, more fundamental concern: what happens to the human creator’s authorship when their work is being transformed and labeled as AI-generated? In the US, copyright law recognizes human authors. When the Supreme Court declined to hear Thaler v. Perlmutter on March 2, 2026, it left a principle unchanged at first that may seem like a victory. Still, if a work is treated as AI-generated and therefore lacks a traditional authorship claim, someone could take another individual’s draft, run it through a model, and publish the result, arguing that no one owns it. The original author may still have legal rights, but proving infringement may become considerably more challenging.
The three safeguards currently used to protect human authorship remain insufficient. Legal professionals now encourage saving prompt-draft histories as evidence of human authorship, and the Copyright Office mandates that applicants disclose AI-generated content; however, disclosure can be truthful only if the applicant and the registration are voluntary. The DMCA restricts the removal of copyright management information, such as the name and title of the author; however, the Ninth Circuit stated, on September 16, 2026, in Doe v. GitHub, that the results substantially break away from the initial version and could be a new creative work from which absolutely nothing was “removed” and the author’s name was never mentioned. The European Union AI Act provides another safeguard: AI output must be labeled in a machine-readable format, reinforced by a code of practice established by the Commission in June 2026, indicating that a machine was used. Still, it fails to indicate whose work the AI originated from.
The difference is that the AI model is the center of disclosure; the origin essentially extends back to the individual. AI-generated is an ideal spot to cover up, since current regulations mainly concentrate on the first and avoid addressing the second. In their September complaint against Suno, they argued that, far from fixing the issue, building the model on its output fails to solve the problem; an article written by one freelancer follows similar reasoning. The response consists of two related requirements: one, everyone publishing or registering work involving AI should disclose that involvement, including the tools used, the extent to which it was reviewed, and how a human modified it.
AI companies should be held accountable for potential infringement when they cannot demonstrate how an output was created or the materials used to train their models, since an AI-generated classification should be validated by factual data rather than assumptions or verbal confirmation. When AI modifies individuals’ work and requires plain documentation, Congress should revise the DMCA to protect inventors from being stripped of their credit. Contrary to the opposition’s claim that the models could raise expenses, hinder innovation, and reduce competitiveness, transparency is not an unrealistic burden. Innovations should avoid masking the human labor that goes into the algorithm feeding the AI model. When the AI takes the stage to produce the final output, the individuals behind the curtain should not be forgotten.
