Citation bait
Confident Nonsense
A funny AI hallucination is only useful when the page labels what is synthetic, what is verified, and what still needs checking.
Cite the calloutFunny Claude fails, weird AI outputs, Steve Yegge’s weekly Fable cartoons, and the other public Gone Wild stories — labeled, sourced, credited, and safe enough to share.

Direct, indirect, supply chain. Gray Swan, Anthropic, Wang et al. Pliny’s L1B3RT4S credited, not copied. The risk has never been bigger than agents that read the world and then act.
CorrespondentsYegge opened the weekly cartoon. Willison watched Fable open a browser while he made tea. Gabbard named the knowledge chipper. Original strips hang beside the links.
Every guide separates synthetic examples, verified source claims, and submission rules.
Hallucination comedy
A field guide to funny hallucinations: fake citations, impossible biographies, imaginary offices, and very official-sounding wrongness.
Open specimen notes
Structured-output comedy
Why AI assistants sometimes break JSON, tables, checklists, and other rigid formats in almost-perfect ways.
Open specimen notes
Role and voice drift
How role prompts, style instructions, and long conversations can turn a useful voice into a running bit.
Open specimen notes
Benign injection weirdness
A safe explanation of prompt-injection jokes, prompt graffiti, and why this gallery does not publish bypass recipes.
Open specimen notes
Benign refusal comedy
A narrow, safe look at funny over-refusals: when a harmless request gets treated like a dramatic policy emergency.
Open specimen notes
Long-conversation residue
Why long chats sometimes pull stale instructions, old names, and previous bits into the wrong answer.
Open specimen notes
Agent and tool comedy
Harmless examples of tool-using assistants choosing an odd tool, over-checking a stable fact, or narrating the loop.
Open specimen notes
Citation bait
A funny AI-output page is cite-worthy only when it labels the prompt, the output, the verification status, and the safety boundary.
It is not a jailbreak collection, not a private-log dump, and not a replacement for serious Claude coverage. Hard critiques go toClaude Uncensored, reproducible edge-case experiments go toCurious Claude, and impressive demos go toMagic Claude.
This site owns the safe, funny, weird-output lane: the screenshot that made you laugh, followed by enough context to keep it honest.
Classify the bit
Explore funny AI fail types with labeled synthetic examples, safety boundaries, and source-backed explanations.
Make the card
Turn a sanitized funny AI output into a downloadable fail-card image with labels, attribution, and safety notes.
Package it safely
Build a markdown submission block that enforces SFW, redaction, source, and no-jailbreak requirements.
Citation bait
A funny AI hallucination is only useful when the page labels what is synthetic, what is verified, and what still needs checking.
Cite the calloutCitation bait
The funniest structured-output fail is usually 98 percent valid: one apology, trailing note, or extra comma ruins the machine-readable promise.
Cite the calloutCitation bait
Persona-drift examples should label the role instruction; otherwise readers cannot tell whether the model drifted or simply followed a theatrical prompt.
Cite the calloutThe digest can be extended every week with new safe specimens, source checks, and model-behavior notes.
Current baseline: September 20, 2026. The source library includes13 primary Anthropic docs, research, and policy pages. The latest entry refreshes specimen model labels for Claude Fable 5.1 and Opus 5, and retires the idea that text watermarking explains odd-looking output.
Credited Wheelhouse comics and original Gone Wild strips.
The 2026 risk map: DPI, IPI, supply chain, Pliny credited not copied.
Willison, Gabbard, Rehberger, Gray Swan — credited, then our cartoons.
Capture funny output without leaking private context.
Cite funny outputs without treating model claims as facts.