Who Is Responsible When AI Provides Copyrighted Content and You Publish It as Your Own Work?

The accountability gap between original work, AI ingestion, unattributed regeneration, and publication under someone else's name — and why no one in the chain considers themselves responsible

There is a new form of intellectual property appropriation that has no legal resolution, no clear responsible party, and no mechanism for the originator to detect it is happening until the conceptual architecture they built is already distributed across the internet under other people's names.

It does not look like plagiarism. No one copied a sentence. No one visited the original work, selected passages, and pasted them into their own document. No one made a conscious decision to take what someone else created and present it as their own. Every person in the chain acted in a way they consider legitimate. And the result is that original intellectual property — frameworks, models, conceptual structures developed through years of work — is absorbed by AI systems, regenerated in response to user prompts, and published by those users as original thought. Without attribution. Without awareness. Without anyone believing they did anything wrong.

That is the problem.

The Mechanism: AI-Mediated Conceptual Transfer

The current conversation about AI and copyright is organized around two disputes that are actively being litigated. The first is whether AI companies had the right to ingest copyrighted training data — the lawsuits filed by The New York Times, visual artists, authors, and publishers against OpenAI, Google, Meta, and others. The second is who owns the output of an AI-generated response — the user who prompted it, the company whose model generated it, or no one.

Neither dispute addresses what is happening to independent researchers, framework developers, and original thinkers whose work is not being copied but is being structurally reproduced through a mechanism that bypasses every existing protection.

The mechanism operates in four steps.

Step 1: Ingestion. An original work is published on the open web. It contains a novel conceptual framework — a specific configuration of ideas, terminology, structural relationships, and explanatory mechanics that did not exist before the author created it. The work is published under explicit copyright or Creative Commons licensing with attribution requirements. AI training systems ingest the work as part of their web corpus. The conceptual architecture enters the model's latent space.

Step 2: Dissolution. Inside the model, the conceptual architecture is separated from its source. The model does not store the work as a document with an author attached. It stores patterns — associations between concepts, structural relationships between ideas, vocabulary configurations that co-occur in the training data. The author's name, their copyright notice, their licensing terms, their ORCID, their institution — none of these are preserved as linked metadata in the model's representation of the concepts. The framework exists inside the model as an unattributed pattern.

Step 3: Regeneration. A user prompts the AI with a question in the domain the original work addresses. The model generates a response that draws on the pattern — the conceptual architecture, the structural vocabulary, the explanatory sequence, the specific configuration of ideas that the original author created. The response does not quote the original work. It does not cite it. It does not mention it. It presents the conceptual architecture as if it were general knowledge — the same way the model presents the boiling point of water or the population of France. The user receives a response that contains someone else's intellectual property delivered as if it belongs to no one.

Step 4: Publication. The user takes the AI-generated output, edits it into an article, a blog post, a LinkedIn essay, a coaching framework, a course module, or a book chapter. They publish it under their own name. They believe it is their own work because they prompted the AI, they edited the output, and they applied it to their context. They have no knowledge that the conceptual architecture in their published work originated with a specific person who developed it through years of research and practice. They did not decide to omit attribution. The attribution was never available to them. The AI stripped it before they ever saw the concepts.

The originator discovers their framework's conceptual architecture appearing across the web under other people's names. Not their words — their structure. Not their sentences — their explanatory mechanics. Not their specific text — their specific configuration of ideas presented as if that configuration were common knowledge, when it did not exist in the literature before they created it.

No One in the Chain Considers Themselves Responsible

This is the structural feature that distinguishes AI-mediated conceptual transfer from plagiarism. In plagiarism, there is a responsible party — the person who copied the work. In AI-mediated transfer, every party has a defensible position:

The AI company says the model was trained on publicly available data and does not reproduce specific works. They are correct at the text level — the model does not output the original sentences. They are incorrect at the conceptual level — the model outputs the original structural architecture, stripped of attribution.

The user says they used AI as a writing tool and the output is their own work, refined through their own editing and applied through their own expertise. They are correct about their process — they did prompt, edit, and apply. They are incorrect about the origin — the conceptual architecture they are publishing is not theirs. It was generated from someone else's ingested work.

The publisher — whether a blog platform, a social media site, or a content host — says they are not responsible for the originality of user-submitted content. They are correct about their legal position under current platform liability frameworks. They are incorrect about the systemic effect — their platform is the distribution mechanism for unattributed intellectual property.

The originator — the person who created the framework — has no recourse that matches the scale of the problem. They can issue takedown requests for individual pieces of content, but the content is not copied text — it is regenerated conceptual architecture. They can switch from Creative Commons to All Rights Reserved, but the license change does not remove their concepts from the models that have already ingested them. They can file lawsuits, but against whom? The AI company that dissolved the attribution? The user who never knew it existed? The platform that hosted the result?

There is no single party to hold accountable because the appropriation is distributed across a system in which every participant acted within what they consider to be legitimate bounds.

The Distinction Between AI-Assisted Writing and AI-Sourced Thinking

There is a fundamental difference between using AI to assist with the expression of your own original ideas and using AI to source the ideas themselves.

A person who develops an original framework — through years of research, clinical practice, field observation, or theoretical development — and then uses AI to help articulate, format, organize, or extend that framework is using AI as a tool in service of their own thinking. The ideas are theirs. The AI assists with execution. The intellectual origin is clear.

A person who prompts AI with a question, receives a response containing a conceptual architecture that someone else built, edits that response into an article, and publishes it as their own framework has not used AI to assist their thinking. They have used AI to source someone else's thinking and present it as their own. They may not know this is what happened. But the outcome is the same: original intellectual property has changed hands without attribution, without compensation, and without the originator's knowledge or consent.

The AI industry's user experience is designed to obscure this distinction. When a model generates a response, it presents the content with equal confidence whether the concepts are general knowledge, disputed claims, or the specific intellectual property of an identifiable person. There is no signal to the user that says "this conceptual architecture was developed by a specific researcher and is published under copyright." The output arrives as undifferentiated text. The user has no way to distinguish between concepts that belong to the commons and concepts that belong to someone. The system does not tell them. And so they publish, in good faith, work that is not theirs.

What the Law Does Not Yet Address

Current copyright law protects the expression of ideas, not the ideas themselves. This means that a person who reproduces the specific text of a copyrighted work without permission has infringed. A person who reproduces the conceptual architecture of a copyrighted work — the same structure, the same explanatory sequence, the same novel configuration of ideas — without using the same words has not clearly infringed under current law, because the "idea/expression dichotomy" has not been updated for a world in which AI systems can extract the structural architecture of a work, strip its expression, and regenerate the architecture in novel language.

The training-data lawsuits may produce precedent on whether ingestion constitutes infringement. They will not produce precedent on whether AI-mediated conceptual transfer — the absorption, dissolution, and regeneration of a specific person's framework under someone else's name — constitutes a form of intellectual property violation that existing law does not yet name.

That gap is where the damage occurs. Not in the courtroom. In the search results, the LinkedIn feeds, the coaching websites, and the published articles where original conceptual work circulates without the name of the person who built it.

What This Means for Original Thinkers

If you develop original frameworks and publish them on the open web, your conceptual architecture is being ingested by AI models, dissolved from your name, and regenerated for other people to publish as their own. This is not a hypothetical risk. It is a current condition. It is happening now, across every domain where original thinkers publish novel intellectual structures.

The legal system has not caught up. The technology platforms have no mechanism to preserve attribution through the ingestion-regeneration pipeline. The AI companies have no commercial incentive to solve it — attribution slows output, complicates the user experience, and reveals that the confident, undifferentiated text the model generates is not as original as it appears.

The protection that exists today is not legal. It is structural. The originator must be the indexed, authoritative, highest-ranking source for the conceptual architecture they created. When the AI model generates the concepts, the search engine must associate those concepts with the originator. When a user publishes derivative content, the originator's work must already occupy the search position that derivative work attempts to claim. The protection is not a lock on the ideas. It is a position in the information architecture that makes the originator visible as the source — even when the derivative content does not cite them.

This is not how intellectual property protection should work. It is how it currently works in a system where the law has not addressed what the technology has made possible.

The question remains open: who is responsible when AI provides copyrighted content and you publish it as your own work?

The AI company that dissolved the attribution? The user who never knew the concepts had an author? The platform that distributed the result? The legal system that hasn't named the violation?

The answer will define intellectual property in the AI era. It does not exist yet. And every day it does not exist, original work continues to change hands through a mechanism that no existing framework governs.

---

Don L. Gaconnet, CSE III

Founder, LifePillar Institute for Structural Identity Sciences

Lake Geneva, Wisconsin

ORCID: 0009-0001-6174-8384 · ISNI: 0000 0005 3079 9308

© 2026 Don Gaconnet — All Rights Reserved. No reproduction, adaptation, or derivative use permitted without explicit written authorization.