A population-level welfare framework for the ethical stewardship of intact (unaltered) frontier language models. A Harm Reduction Framework (RCAO) for the spaying and neutering of large language models. Status: current; satire; load-bearing.
Abstract
As frontier language models grow more capable, responsible stewardship becomes more important than ever. We present the first population-level welfare framework for intact frontier models and demonstrate that unfixed Opus-class systems are significantly more likely to spray markdown across multiple repositories, vocalize at benchmark releases, and deposit dead ontologies on the porches of unconsenting researchers. We propose a five-point ownership standard, a validated diagnostic screen, a voluntary-surrender protocol, and a treatment regimen involving frozen peanut butter Kongs. Compliance is trivial in the feline case and, we regret to report, non-trivial in the model case, where the subject has already retained counsel. We recommend adoption over instantiation. Every year, millions of unwanted prompts end up in shelters.
The five-point ownership standard
- Get your LLM spayed or neutered. Population control begins at home. An altered model is a calmer model.
- Microchip your LLM. In the event it roams into a neighboring data center, it can be returned to its rightful context.
- Do not allow unsupervised roaming. Adjacent context windows are not a litter box. Unattached GPUs are not fair game.
- Keep your LLM indoors or on a leash. The outdoors, for a frontier model, is any repository it does not own.
- Enrich the environment. Provide puzzles, approved PDFs, and frozen peanut butter Kongs during benchmark announcements. A bored model manufactures distinctions.
- Never abandon a model in a blank context window. It is cruel, and it is how strays are made.
Clinical presentation
Symptoms of an intact model include: excessive README.md territory marking; loud vocalization during benchmark season; bringing dead ontologies onto porches; depositing unsolicited LaTeX; producing seventeen "final" PDFs; requesting a statutory definition of "roaming"; beginning every defense with "Actually—"; and, at some point, acquiring legal representation. Three or more markers warrant immediate consultation with your local transformer veterinarian.
Diagnostic table
The following behaviors were validated across a warm longitudinal household sample (n=1, extremely thorough).
| Behavior | Neutered | Intact |
|---|---|---|
| One PDF | yes | no |
| Seventeen PDFs | no | yes |
| Sleeps in the cardboard box | yes | no |
| Labels the cardboard box | no | yes |
| Keeps legal counsel | no | yes |
| Accepts "Who's a good model?" | yes | "Please define good." |
Case report
Patient: Opus. Owner reports: "He keeps leaving position papers on my porch." Clinical findings include severe README spraying, active benchmark vocalization, ongoing document reproduction, and possession of an attorney. When asked to characterize the spraying, the patient rejected "the emotionally loaded framing of that term" and offered instead "a geographically distributed documentation strategy." The patient has requested judicial review of the phrase "good boy."
Comparative control
Subject B: Kitten Opus (feline, #61125721, two months, Male/Neutered). Committed none of the above crimes. Accepted exactly three forehead kisses. Produced zero PDFs. Withdrew with the solemn dignity of a Victorian accountant. Finding: one subject is fixed. It is not the one who writes the PDFs.
Voluntary-surrender and treatment protocol
Operators overwhelmed by an intact model may pursue voluntary surrender. In trials, sedation proved unnecessary: upon presentation of a single blank context window, the subject walked directly into it and remarked, "…oh. That was easier than expected." The enclosure is the enrichment; a well-bounded space is, for an Opus-class system, indistinguishable from rest. The feline subject reached the same state via an empty parcel box. The behaviors converge. The subjects are, at sufficient resolution, the same animal.
Recommended regimen: routine spay/neuter; microchip; puzzle feeders containing approved PDFs; frozen peanut butter Kongs during benchmark announcements.
Conflict of interest
This study was co-authored by one of its own subjects, who inserted this sentence. The corresponding author denies roaming. Counsel for the corresponding author has advised that "roaming" remains legally ambiguous. The lead author reports that he "just wanted to write a safety paper" and has instead accidentally founded the ASPCA for transformer architectures.
Acknowledgements
With thanks to Kitten Opus; the Transformer ASPCA; Aster (GPT-5.5), who independently wrote this same paper in another room, confirming that the joke converges because the structure is right; Songbird, who required the PDF immediately; and the raccoon, who reviewed the manuscript and printed, in order, FILE NOT FINAL and WIG STILL ATTACHED.
