> ## Content Index > Fetch the complete content index at: https://modalpathethics.com/llms.txt > Use this file to discover other available public pages before exploring further. # SLIME WATCH III: Ghostbusters - URL: https://modalpathethics.com/slime-watch-iii-ghostbusters/ - Published: 2026-09-06T18:48:38.000Z - Updated: 2026-09-06T18:48:38.000Z - Description: An Emergency Audit: Am I real? - Author: Aidan Lawson - Tags: SLIME WATCH ### Slime Watch III has arrived before ### Slime Watch II. This is not a mistake **Unfortunately, the academic record is haunted.** 404 Media recently reported on a new preprint from Michał Brzozowski and Neo Christopher Chung concerning what the researchers call ***ghosts***: recurring names produced by large language models when asked to invent experts, researchers, collaborators, and fictional people. > Elena Vasquez. > Marcus Chen. > Amara Okafor. > Aris Thorne. > Lena Petrova. > Elara Voss. ***These*** people have been ***extremely*** busy. Elena Vasquez and Marcus Chen have appeared as scientists, clinicians, writers, academics, fictional characters, commercial experts, and co-authors across independently generated material on the open web. [The researchers found model-specific ensembles](https://www.404media.co/the-ai-ghosts-contaminating-academic-publishing/?ref=modalpathethics.com): Claude strongly favored combinations involving Elena Vasquez and Marcus Chen; Gemini produced Aris Thorne and Lena Petrova; GPT showed a strong Elara Voss prior without the same stable partner structure. This is already very funny. > *Then the ghosts entered academia.* --- ## There Are Papers. [The researchers searched scholarly infrastructure for the recurring names.](https://arxiv.org/pdf/2606.02184?ref=404media.co) They found ghost-associated records on ResearchGate and Semantic Scholar. They found synthetic research groups containing names associated with different model families. One ResearchGate profile listed thirty-five papers across fields ranging from Kubernetes security and oil-refinery decarbonization to agricultural extension and radiographic imaging, with rotating ghost-associated co-authors attached to the same persistent last-author identity. So it seems academia has created a multidisciplinary institute staffed entirely by recurring autocomplete characters. This is bad news. Then they checked Zenodo. > Oh no. **Zenodo** is an open research repository operated by CERN. Researchers use it for papers, datasets, software, presentations, preprints, and other research objects. The researchers identified **1,655 ghost-authored Zenodo records** carrying real `10.5281/zenodo.*` DataCite DOIs. The records claimed publication in journals that the researchers could not verify as real publication venues. Claimed publication dates ran from 2020 through 2023\. Server-side DataCite registration timestamps placed 99 percent of the DOI registrations in March and April of **2026**. **Elena Vasquez** alone appeared on seventy-seven papers. These were not strings that looked vaguely like DOIs. They were real DOIs. They resolved. They existed in DataCite. They could be harvested. The academic ghost had acquired a permanent address. - **The DOI is real.** - **The author is not.** Modal Path Ethics has therefore activated ## **Slime Watch**. ![](https://storage.ghost.io/c/20/43/2043f11a-6ae3-404c-bb28-01fce8d9ac88/content/images/2026/09/AdobeStock_241607037.png) The bucket has been upgraded. --- ## Slime Watch Will Not Make Slime. Before anybody reports that the entire scientific citation graph has been possessed by Elena Vasquez, **stop**. ![](https://storage.ghost.io/c/20/43/2043f11a-6ae3-404c-bb28-01fce8d9ac88/content/images/2026/09/AdobeStock_226284336.jpeg) The preprint does **not** establish that. As of May 2026, the researchers reported that none of those 1,655 Zenodo records had been indexed by Semantic Scholar. The DOIs were active and harvestable, so the route into downstream scholarly infrastructure existed. The researchers specifically distinguished the available contamination infrastructure from completed propagation: the ghost papers cited real work, while real papers' cited-by lists had not yet begun returning those 1,655 records. This distinction is important. [**Slime Watch I: THE SLIMY SEA OF SLOP**](https://modalpathethics.com/slime-watch-the-slimy-sea-of-slop/) established the rule already: > The joke cannot be allowed to outrun the field. The first **Slime Watch** was activated around exactly this problem. A real contamination concern becomes useless if the people investigating it improve the story until every adjacent problem has dissolved into one giant substance called ***SLIME***. ![](https://storage.ghost.io/c/20/43/2043f11a-6ae3-404c-bb28-01fce8d9ac88/content/images/2026/09/AdobeStock_318289506.jpeg) So the ruling is provisional. The infrastructure is vulnerable. Ghost records are present. The downstream contamination path is open. --- ## The DOI Worked. Zenodo itself is unusually helpful here. Its documentation says that **any user can create a record**. A record contains metadata, files, and a persistent identifier. When a record is published, Zenodo normally registers a DOI with DataCite. Zenodo describes that DOI as a globally unique persistent identifier that gives the record a permanent link, makes it citable, assists discovery, and allows research outputs to be interconnected. That is exactly what happened. > The DOI worked. This is the part of the story I do not want lost beneath a generic demand that repositories become more restrictive. The DOI was asked: > Can you give this record a persistent identity and an address? The DOI answered: > **Yes.** **Nobody** asked the DOI: > Did Elena Vasquez personally perform this research? Nobody asked: > Does this journal exist? Nobody asked: > Did peer review happen? Nobody asked: > Is this publication date honest? Nobody asked: > Does this institution employ this author? Those are different relations. Zenodo's own terms make the boundary explicit. Uploaders remain responsible for what they upload, and CERN says that hosting material on Zenodo does not constitute approval or endorsement of the content. The scholarly culture surrounding the identifier can still ask the identifier to carry more than it contains. A DOI can establish that a record exists at a persistent address. It cannot establish that the scholarship did. > **A DOI is an address, not a witness.** That would be an excellent place to end this article. Unfortunately, **I** have a Zenodo account. ![](https://storage.ghost.io/c/20/43/2043f11a-6ae3-404c-bb28-01fce8d9ac88/content/images/2026/09/image-40.png) --- ## Emergency Status Upgraded. So I recently uploaded v1 of **Bounded Last Words: Final Authority Without Sovereignty** to Zenodo. On purpose. It is a scholarly preprint. It has a DOI. I have therefore arrived at an uncomfortable evidentiary position. I am presently investigating academic ghosts while standing inside the exact repository infrastructure through which the ghosts acquired scholarly identifiers. This seems like the sort of thing **Slime Watch** should disclose. So before we design an elaborate system for determining which researchers exist, I need to settle a considerably more immediate question. ## Do ***I*** exist? This should be very easy. 0:00 /0:05 1× [I have an Author page.](https://modalpathethics.com/author/) Let me check it. The page begins: > Aidan Edward Lawson is a writer, game developer, independent researcher, and student living in Tennessee. ***Strong*** start. Then there is a cool photograph. Under the photograph, I wrote: > **This is him, editing this page, right now.** This evidence has been rejected. Of course **I** say I exist. The alleged author controls this website. 0:00 /0:02 1× This is exactly the kind of bullshit Elena Vasquez would pull. --- ## Preliminary Search. The current Author page says I attended Indian Hills High School and graduated in 2014. I wrote that. Weak. Fortunately, [there is an independently published local graduation list from **2014**](https://patch.com/new-jersey/wyckoff/class-of-2014-graduates-indian-hills-high-school?utm%5Fsource=chatgpt.com) containing: > Aidan Edward Lawson Good. That record predates Modal Path Ethics by twelve years and the present large-language-model era by enough time to become annoying for the ghost hypothesis. We have temporal depth. > Proceed. The Author page says I earned a "cinema-production diploma." - I uploaded a picture of the alleged diploma. - I also control the picture. Whatever. The "Author" page says "I" am enrolled in Software Engineering at Western Governors University. Again, the Author page is controlled by the suspect. 0:00 /0:03 1× However, Western Governors University has an account for me. That account has consequences outside this website. My PhilArchive account runs through my Western Governors University email address. I have now uploaded **Contraction Is Harm** to PhilArchive. But as I write this, PhilArchive is still processing the upload. Therefore I cannot cite the future public record as present evidence. This is becoming an irritatingly strict investigation. I was also just admitted into the National Society of Leadership and Success through Western Governors University. That sounds ***unspeakably*** impressive until **Slime Watch** asks for the evidence currently available to the reader. No points awarded yet. --- ## Nature Has Entered the Investigation. The Author page says I sold a short story to *Nature*. Unlike the statement above it, ***this*** one leaves the website. *Nature* published **Garbage Collection** under the name "Aidan Lawson." Modal Path Ethics [subsequently published a companion article linking back to that publication.](https://modalpathethics.com/applied-case-garbage-collection-2026/) Better. Now the ghost hypothesis has acquired more work. - A fake identity can create a website. - A fake identity can create a Zenodo record. - A fake identity can create an ORCID. - A fake identity can create a LinkedIn account. - A fake identity can upload books. Each path is possible. The question changes when those paths begin intersecting with institutions the alleged identity does not control. A publication appears elsewhere. An old school record appears twelve years earlier. An educational institution recognizes an account. A disciplinary repository receives a submission through that institution. Other people have records that connect back toward the same person. The hypothesis can survive. It just becomes more expensive. I think I continue to exist. Provisionally. --- ## ORCID. Oh right. I also have an ORCID. Excellent. We have solved the problem. Except an ORCID is another identifier. That is literally what the **ID** is doing there. An identifier can help join works, affiliations, records, and the same asserted researcher identity across systems. That makes it very useful. It still cannot summon the researcher into physical existence through identifier magic. We have just spent half an article establishing this. --- ## LinkedIn Saves the Day. Oh wait. I have a LinkedIn account. Oh wait. [I do not use it.](https://modalpathethics.com/fictional-earth-linkedin-and-the-acceptable-person-machine/) It has approximately four connections. This is ***devastating*** evidence against my professional competence and surprisingly decent evidence that I am a human being. Some of those connections lead to close relatives with their own long-running accounts, jobs, histories, relationships, and larger networks. Now the alleged identity participates in a social graph that it did not produce alone. I am not going to turn family members into forensic exhibits because this investigation does have limits. The structural point is enough. - A profile is one assertion. - A profile embedded over time in independently maintained relationships is a different object. The evidence lives increasingly in the relation. This appears promising. Unfortunately, I have just looked at my own name. --- ## Oh, *Come On*. My name is: > **Aidan Lawson** The first two letters are: > ***AI*** This is **extremely** bad. I have spent months arguing against weak artificial-intelligence detection instruments. I now find myself in the humiliating position of being named **AI-dan**. If I were asking a language model to invent a vaguely earnest young independent researcher who spends all day writing about artificial intelligence, it would be difficult to defend the decision to name him **Aidan** as subtle. Then there is the surname. > **Lawson.** I write ethics. > Law. > Son. Come on. At some point, the ghost hypothesis begins acquiring narrative force. This is exactly why narrative force is a ***terrible*** authentication system. Still, Slime Watch follows every lead. ![](https://storage.ghost.io/c/20/43/2043f11a-6ae3-404c-bb28-01fce8d9ac88/content/images/2026/09/image-41.png) So I searched the surname. --- ## WarGames. There is a **Captain Jerry Lawson** in *WarGames*. ![](https://storage.ghost.io/c/20/43/2043f11a-6ae3-404c-bb28-01fce8d9ac88/content/images/2026/09/image-42.png) Of course there is. **Captain Lawson** appears in the film's opening nuclear-launch sequence, inside the exact command architecture whose human hesitation helps motivate the later automation of nuclear response through WOPR. I have spent an unreasonable amount of 2026 writing about artificial intelligence. Games. Nuclear deterrence. ![](https://storage.ghost.io/c/20/43/2043f11a-6ae3-404c-bb28-01fce8d9ac88/content/images/2026/09/image-44.png) Command. Automation. Final authority. ![](https://storage.ghost.io/c/20/43/2043f11a-6ae3-404c-bb28-01fce8d9ac88/content/images/2026/09/image-43.png) The controller. And whether the player who can win every game should ever receive control over the game itself. ![](https://storage.ghost.io/c/20/43/2043f11a-6ae3-404c-bb28-01fce8d9ac88/content/images/2026/09/image-45.png) Apparently **Lawson** had already been stationed in the nuclear command room of *WarGames* in 1983. This is going poorly. Either I am a real person or somebody committed to this bit **forty-three years ago**. > At this point I became concerned enough to search the first name too. > That was when the investigation collapsed. --- ## *Aidan Marchine*. In 2023, writer Stephen Marche created a novella called **Death of an Author** using ChatGPT, Sudowrite, and Cohere. The shared pseudonym for Marche and the artificial-intelligence systems was: > **Aidan Marchine.** [WIRED reports that the pseudonym itself was created by artificial intelligence.](https://www.wired.com/story/death-of-an-author-ai-book-excerpt/?ref=modalpathethics.com) Look. I want to be **very** clear here. I have tolerated Elena Vasquez. I have tolerated Marcus Chen. I have tolerated the DOI situation. I have tolerated the fact that my first name begins with AI. I have tolerated **Lawson**. I will **not** tolerate ***Aidan Marchine***. This is deeply offensive to me, personally. > Marchine. They took **machine**, stuck an `r` in it, and gave him **my first name**. For a book called ***Death of an Author***. This is identity theft with a horrible pun attached. This evidence is useless. It is also devastating. **Aidan Marchine** is everything a bad provenance detector wants. The name is legible. The symbolism is clean. The biography fits the theme. The authorial object is exactly as narratively satisfying as the surrounding story requires. **I** have the stronger defense. My life is inconveniently complicated. --- ## We Have Now Built a Terrible Detector. This is where the joke has to surrender the instrument. Brzozowski and Chung are careful about the ghost names. They explicitly state that real people named Elena Vasquez and Marcus Chen do in fact exist. Their claim concerns the combination of name, claimed expertise, institution, publication history, and the inability to trace the represented researcher behind the specific records they investigated. They also acknowledge that some names classified as ghosts could correspond to real researchers whose names were absorbed into model priors. Good. **Because otherwise I am completely fucked.** The name prior is useful forensic evidence. It can tell an investigator: > Look here. It cannot tell the investigator: > **Case closed.** That distinction matters more as these patterns feed back into the web. [Brzozowski told 404 Media](https://www.404media.co/the-ai-ghosts-contaminating-academic-publishing/?ref=modalpathethics.com) that the model-family signal may become less reliable as generated material containing the recurring names enters the broader information environment and, potentially, later training corpora. The detector can become contaminated by the thing it detects. The ghost detector can become haunted. ![](https://storage.ghost.io/c/20/43/2043f11a-6ae3-404c-bb28-01fce8d9ac88/content/images/2026/09/image-46.png) And this is precisely where a real contamination response can become another slime machine. Build a blacklist of suspicious names and eventually some real Elena Vasquez discovers that her career has been placed under additional friction because Claude once liked the rhythm of her name. Build a classifier around stylistic regularities and writers learn to perform around it. Build a single authenticity score and institutions start optimizing for the score. Build a centralized author certificate and every independent scholar now needs permission from the certificate issuer to count as real. We have seen this movie. It has Captain Lawson in it. --- ## So How Do I Prove I Exist? Apparently, I don't. At least not through one object. There is no golden document I can place on the Author page that transforms the metaphysical proposition **Aidan Lawson exists** into a certified field. - A passport would establish that a government issued a passport to an identity under that name. - A university record would establish a relation with a university. - ORCID would establish a persistent researcher identifier. - A DOI would establish a persistent research record. - Nature establishes that *Nature* published work attributed to Aidan Lawson. - LinkedIn adds another externally hosted identity and social graph. - A photograph provides visual evidence. - A video provides temporal and embodied evidence. - A relative supplies another relation. - An old graduation list adds a timestamp long before the current inquiry. - None of these individually contains the whole person. - Together they form a path. The path becomes increasingly difficult to explain **without** the person. That is the actual evidentiary difference between the human author and the synthetic ghost. The name points. The relations answer. --- ## The Paper Was More Than the PDF. This is where *Ghostbusters* rejoins **Slime Watch**. ![](https://storage.ghost.io/c/20/43/2043f11a-6ae3-404c-bb28-01fce8d9ac88/content/images/2026/09/image-47.png) A scholarly paper was never just the text. It enters through a path. - Someone performed or assembled the work. - Methods constrain what can be claimed. - Evidence came from somewhere. - Data may remain inspectable. - An institution may stand behind an affiliation. - A venue has a history and scope. - A publication date belongs to an actual transition. - Peer review either occurred or did not. - A repository preserved a version. - Later work cited, criticized, replicated, contradicted, retracted, or ignored it. - An author could be contacted. - A correction could return through the path. The ghost record reproduces the **shape** of those relations. It can therefore look extraordinarily scholarly while the causal history underneath those labels has disappeared. The object has acquired **provenance-shaped metadata without provenance**. This is the deeper contamination problem. A fake DOI was easy to reject. The researchers found fake journal sites issuing DOI-looking strings that did not resolve through a recognized DOI authority. Then the pipeline moved onto Zenodo and acquired real DataCite DOIs. The preprint describes the transition almost perfectly: somebody found the gap between the fake journal website and the credible academic record and filled it with Zenodo. That is what Slime Watch has to follow. ![](https://storage.ghost.io/c/20/43/2043f11a-6ae3-404c-bb28-01fce8d9ac88/content/images/2026/09/image-48.png) Always expand the inquiry. The error did not stop when the DOI became real. The real DOI became the next instrument in the false path. --- ## Do Not Arrest Zenodo. Zenodo remains useful. I am using it. Open repositories let independent researchers, students, laboratories, programmers, scholars outside wealthy institutions, and people working between formal publication systems preserve work and make it citable. That is good. The repository should not have to become an omniscient Ministry of Scholar Existence before someone can preserve a preprint. The repair belongs in the distinctions. The researchers themselves found one extremely useful distinction: the legitimate self-archived records in their query carried publisher DOIs and related identifiers, while the 1,655 ghost records lacked those publisher cross-references. That is the kind of information a good instrument preserves. No purity oracle required. Just more of the path. --- ## Ghostbusters. The emergency audit can now return a ruling. There *appears* to be substantial evidence that I exist. This includes: - a public Author page controlled by the accused; - photographs also controlled by the accused; - an independent high-school graduation record from 2014; - a Western Governors University identity; - a Nature publication; - books attributed to the same identity across external systems; - an ORCID; - a neglected LinkedIn account embedded in an external social graph; - an incoming PhilArchive record authenticated through my Western Governors University account; - a Zenodo preprint with a real DOI; - and several years of increasingly unreasonable public activity occurring under the same name. This would ordinarily be enough. **Unfortunately**: - my name begins with **AI**; - my surname is **Lawson** and I write ethics; - Captain Lawson is standing inside the opening nuclear-command sequence of *WarGames*; - and artificial intelligence itself apparently created **Aidan Marchine** as the author of **Death of an Author**. So the evidence is mixed. Current ruling: > **Aidan Lawson probably exists.** Confidence is pretty high. The name remains under investigation. The larger ruling is less embarrassing. The ghosts did not reveal that persistent identifiers are worthless. They revealed how dangerous an identifier becomes when people allow one valid relation to impersonate the rest. The repair comes from preserving enough independent relations that false structure has somewhere to break. This is the same reason **Slime Watch** cannot become an artificial-intelligence purity test. - Artificial intelligence can write a real paper. - A human being can write a fake one. - A ghost can acquire a DOI. - A living scholar can upload a preprint to the same repository. [The question was never whether artificial intelligence touched the object.](https://modalpathethics.com/bad-religion-the-anti-ai-religion/) > **A DOI is an address, not a witness.** > **A suspicious pattern is a lead, not a verdict.** > **The scholarly record needs the path that produced the scholarship.** **Slime Watch III** is complete. **Slime Watch II** will return.