The failure mode of ungoverned AI output. Impressive-looking, generic, and indistinguishable from every other model output.
AI slop is the failure mode of ungoverned AI output. It is the category of content that is impressive at a glance, generic at inspection, and indistinguishable from every other model's output on the same prompt. It is what happens when a model is asked to produce without constraint.
The visual signature on the web: glowing gradient buttons, Inter font, center-aligned hero sections with blurred radial backdrops, 3D card flips, neon on black, impossibly smooth animations, stock-photo-style AI-generated images. Every AI-generated site looks the same because every model is optimizing the same objective: impress the person evaluating the output. The specific constraints that would make the output distinctive are not in the prompt, so they do not show up in the result.
The linguistic signature in text: setup paragraphs before the point, hedging qualifiers, corporate buzzwords, em dashes instead of structural breaks, summaries of what was just said, meta-commentary about the process, gradual endings. The text feels smooth because every micro-decision was optimized for smoothness, and the smoothness is what makes it forgettable.
Arkeus treats slop as the baseline failure mode, not an occasional mistake. The brand guidelines exist to resist it. The voice rules in corrections.md exist to resist it. The refraction rule exists to resist it. Every file is a specific constraint telling the model which direction to break out of the smooth average.
The test: if the output could have been generated by an assistant that had never read Ryan's kernel files, the output is slop. Individualism is the product. Average is the failure mode. When the system produces slop, that is itself a correction event — the kernel was not loaded correctly, the domain was not routed, or the rules are drifting.
Slop is not a style preference. It is a measurable outcome of ungoverned optimization.