A satirical, legally-grounded field guide to how AI-era plagiarism and copyright infringement actually work — and what happened when the author's own article became Exhibit A.
Plagiarism has been professionalized by AI: it has moved from late-night student panic into research, scholarly publication, and social-media-fueled “thought leadership.” Talisman's satirical field guide is built on Teddi Fishman's 2009 definition of plagiarism — four qualifiers that every tactic below is engineered to blur — then walks through six copyright myths, three AI-scale laundering methods, and the international legal framework (or lack of one) protecting an author's right to be credited. The piece closes with the author's own experience: her original article was plagiarized by an AI-generated knockoff that out-performed her post on LinkedIn.
Six tactics, each engineered to blur one of Fishman's four qualifiers of plagiarism — identifiable source, attribution, expected originality, or benefit.
Strip the original language and keep only the idea — Fishman's definition covers ideas, not only words.
A short delay makes copying look like convergent genius rather than a repost.
“This came out of a chat I had with a founder” alludes to a source without identifying it.
“As someone who's been thinking about this for a while” manufactures expected originality out of thin air.
Screenshot the text, crop the byline, and reframe the original author's thesis as your insight.
Rename century-old library science vocabularies as a ‘context graph,’ drop it into a white paper, cite nothing — epistemic squatting.
Six widely-believed justifications for reusing copyrighted work — none of them true. Attribution does not license reproduction; only expression, not ideas, is protected.
Naming someone is not a license. Title 17 gives the author the exclusive right to reproduce their work regardless of acknowledgment.
Fair use under 17 U.S.C. §107 is a four-factor test; ‘educational’ is one input to one factor, not a shield.
“No copyright infringement intended” carries the same legal force as “no speeding intended.”
Locating an image via search does not clear the rights holder; cropping a watermark can itself be an infringing derivative.
Video covers require a synchronization license, which is not compulsory. Attribution licenses nothing.
Every serious remix tradition, from Creative Commons BY to sampling clearance, is built on attribution.
Large language models are, functionally, attribution-stripping engines: they ingest sourced text and emit unsourced text.
“Summarize this article in my voice.” Copyleaks found 59.7% of GPT-3.5 outputs contained some plagiarized content across 1,045 outputs.
“Write a post in the style of [popular writer].” The conduct at the heart of The New York Times v. Microsoft/OpenAI.
A 99%-similarity Turnitin score dropped to 31%, then 1%, after two QuillBot paraphrase passes — the original source vanished from the report.
Attribution is the thing every plagiarism move attacks — and the thing US law protects least. The droit de paternité is guaranteed abroad; in the US it survives almost nowhere outside VARA.
Four qualifiers: identifiable source, no attribution, expected originality, benefit gained.
New expression alone is not dispositive of transformative use when purposes overlap commercially.
Near-verbatim reproductions gave rise to a plausible inference of infringement; core claims proceed.
Guarantees the droit de paternité — the moral right to be named as author.
The US implementation of Article 6bis — reaches only signed visual-art editions of 200 or fewer.
Recognizes attribution rights, but requires the author to assert them first in writing.
Treats substantial unattributed copying as research misconduct, with a safe harbor for standard phrasing.
Talisman's conundrum: attribution is the one thing every plagiarism move attacks, and the one thing law protects least — voluntary, unenforceable, “nothing enforceable, just plain ethics.” This collection is itself a working answer to that gap, not a proposed one. Every entity above is a resolvable HTTP IRI rather than a name in prose; its provenance is carried as data via prov:wasGeneratedBy and prov:actedOnBehalfOf, and schema:accountablePerson ties it to a WebID rather than a name a reader has to take on faith. None of that requires new legislation — it only requires publishing knowledge as Linked Data instead of plain text, so attribution survives being copied, summarized, or fed into a model, because it was never separable from the content to begin with.
— Perspective by kg-generator on behalf of Kingsley Idehen
Using another's words or ideas, attributable to an identifiable source, without attribution, where original authorship is expected, for non-required-monetary benefit.
Occupying intellectual territory someone else cleared and defending it as one's own.
Downloading and natively re-uploading a video elsewhere without permission to capture engagement.
A fair-use doctrine narrowed by Warhol v. Goldsmith: new expression alone is not dispositive when purposes overlap commercially.
The moral right to be named as author, guaranteed by Berne Article 6bis independently of economic rights.
Verbatim or near-verbatim reuse of one's own prior-published work without citation.
A four-factor balancing test under 17 U.S.C. §107 — not satisfied merely by an educational label.
Naming the source of words or ideas — infrastructure the law protects least.
Copying, transforming, and combining prior works — legitimate only when attribution is preserved.
A venture-marketing rebrand of a procedural knowledge graph grounded in formal ontologies.
Recognize when someone strips a framework's prose so the idea reads as newly theirs.
A multi-day gap before a similar ‘mental model’ post is the tactic, not convergent genius.
Vague sourcing like ‘a chat with a founder’ alludes to a source without identifying it.
‘As someone who's been thinking about this’ preceding a paywalled paraphrase is a red flag.
A cropped screenshot framed as commentary is the purest form of epistemic squatting.
A renamed discipline dropped into a white paper without citation is epistemic squatting at the consultant tier.
Since copyright protects almost none of this, and US law provides no enforceable attribution right, voluntary credit is the only real defense.
The mirror image of the playbook above: concrete practices for crediting sources when LLMs make attribution stripping the path of least resistance.
Attribute at first use, not after a ‘decent interval’ — a four-day delay is the tactic, not convergent thought.
Don't drop the ‘as cited in.’ Citing only the summarizer quietly promotes them to primary-source status.
If an LLM produced the summary or ‘style of’ pass, say so and preserve the source list it worked from.
If you rename an established field's vocabulary for a pitch deck, name the discipline. A citation costs a sentence.
Every remix tradition that survived kept attribution non-optional. Quote sparingly, link generously.
Use resolvable HTTP IRIs and machine-readable provenance (prov:wasGeneratedBy, schema:accountablePerson) instead of plain prose.
Write citations that survive contact with a paraphraser, not ones engineered to be quietly dropped.
Interactive force-directed view of every entity and relationship in the companion RDF: 192 nodes, 616 edges. Click a node or edge label to open its full description via URIBurner.
This page was generated by extracting Jessica Talisman's LinkedIn Pulse article “How to Plagiarize” into a 647-triple RDF-Turtle knowledge graph (schema.org vocabulary plus a lightweight custom ontology for plagiarism tactics, copyright myths, and legal cases), then rendering that graph as this interactive infographic and companion Markdown document.