About
What is ENAIH?
The Encyclopedia of AI Hallucinations is a community-maintained database of real, reproducible AI hallucinations — confident-but-wrong (or misleading) outputs produced by real, named AI systems. ENAIH aims to be the OEIS of AI failures: a permanent, citable, structured record. Just as the On-Line Encyclopedia of Integer Sequences became the go-to reference for integer sequences, ENAIH aims to be the go-to reference for documented AI hallucinations. Useful for researchers, journalists, AI developers, and anyone who wants calibrated intuition about when to trust AI.
Our vision
ENAIH started because hallucinations were scattered across social media, screenshots, and papers with no central place to find, cite, or browse them. The long-term vision:
- A citable database researchers can reference in papers.
- A resource journalists can link to for concrete examples.
- A dataset AI companies can use to improve training.
- A place anyone can visit to understand AI failure modes.
Who we are
- Rudra Jadhav — Founder
- Leads product vision and research direction. OTIS alumnus, Non-Trivial research program graduate. LinkedIn
- Warren Woolf — Founder
- Leads technical development and infrastructure. Incoming student at Stanford University. LinkedIn
Frequently asked questions
Answers to the questions we hear most often. Still stuck? See the submission guide or get in touch.
- What counts as an AI hallucination — and what should I submit?
- We use "hallucination" broadly — it covers more than just made-up facts. Anything where a real, named AI system fails in a documentable way is fair game: fabricated citations, confidently wrong facts or math, invented code and APIs, temporal confusion, ignored instructions, and misleading or overconfident answers, as well as adjacent failure modes like spiraling, looping, or thrashing (the model degenerating into repetition or runaway tangents). Browse the categories to see the full range. Whatever the type, the best submissions are reproducible: include the exact prompt (or full conversation), the model's response, and which AI model and version produced it, so others can verify and cite it. You can capture a single prompt-and-response pair or a multi-turn conversation. See the submission guide for the structured fields reviewers look for.
- Do I need an account to submit, and can my entry be anonymous?
- Yes, submitting requires an account — it lets you track your submissions, respond to reviewer messages, and edit your entries. You can sign up with an email address (we send a 6-digit verification code) or with Google. If you'd rather not have your name attached publicly, you can mark a submission as anonymous when you submit it; your account stays private and the public entry won't show who filed it.
- How are submissions reviewed, and how long does it take?
- When you submit, your entry goes public right away as pending
review — reachable by its link or A-number but hidden from the
default listings until staff vet it. Every submission gets its permanent
A-number (like
A000042) at this point, so you can cite it immediately. From there it climbs a trust ladder. Staff first confirm it's a genuine, reproducible hallucination and give it a category; it becomes pending acceptance — still hidden from the default listings. Next they try to reproduce it themselves: if they succeed it becomes active — the top tier — and appears in the normal listings; if they can't, it's rejected and kept as a reported sighting. (Link submissions stop at pending acceptance, since staff can't re-run someone else's session.) You can message reviewers from your dashboard at any stage, and you'll be notified when the status changes. We're a small team, so timing varies — there's no fixed turnaround. Submissions that aren't genuine are rejected with a reason, and you can revise and resubmit.
Contact
Reach the maintainers at [email protected]. One inbox covers everything below — just say which it is in the subject line so we can route it quickly.
Prefer to chat? Join our Discord server to talk with the maintainers and community.
- Bug reports
- Something on the site is broken, looks wrong, or behaves oddly. Tell us what you did, what you expected, and what happened (a screenshot and the page URL help a lot).
- Patches & corrections
- An entry is out of date — the model no longer reproduces the hallucination, or a detail is wrong. Include the entry's A-number (or its URL, if it doesn't have one yet) and what should change.
- Problems with a submission
- An entry looks fabricated, miscategorized, duplicated, or otherwise shouldn't be live. Send the A-number or entry URL and what's wrong.
- Suggestions
- Ideas for features, categories, or anything that would make ENAIH more useful. We read all of them.
- Staff applications
- Want to help review the submission queue? Tell us a bit about yourself and why you'd be a good reviewer.
For submitting a hallucination, use the submission form instead — it captures the structured fields reviewers need.
Terms and privacy
See our terms of service and privacy policy.
Source code
The site is open source: github.com/warrenwoolf/encyclopedia-of-ai-hallucinations.
Site built primarily by Claude Opus 4.7/4.8.