hallucinator

Detect fake citations in academic papers. Extract references from PDFs and validate them against 10+ academic databases. If a reference doesn't exist anywhere, it's probably hallucinated.

GitHub Documentation Manifesto

Install

# TUI (recommended — interactive terminal UI)
curl -sSf https://hallucinator.science/install.sh | sh
# CLI
curl -sSf https://hallucinator.science/install-cli.sh | sh
# Python
pip install hallucinator

Pre-built binaries for Linux, macOS, and Windows. Python wheels for 3.12+.

Demo

Hallucinator TUI demo showing reference validation in progress

How it works

1

Extract text from PDF and locate the references section

2

Parse each reference — extract titles and authors from IEEE, ACM, USENIX, AAAI, and other formats

3

Query all databases in parallel — 4 references at a time, all databases simultaneously

4

Report results — verified, author mismatch, not found, or retracted

Databases checked

CrossRefDOIs, journals, conferences
arXivCS, physics, math preprints
DBLPCS bibliography
Semantic Scholar200M+ papers
ACL AnthologyComputational linguistics
Europe PMCLife sciences, 42M+ abstracts
PubMedBiomedical literature
OpenAlex250M+ works
GovInfoUS federal documents
Web SearchSearxNG fallback

Result types

Verified Found in a database with matching authors. The reference exists.
Mismatch Title found but with different authors. Worth investigating.
Not Found Not in any database. Potential hallucination.
Retracted Paper exists but has been retracted. Flagged via CrossRef.

Why this exists

Academia is under attack from AI-generated slop — fake citations, fabricated papers, LLM-written reviews.

The November 2025 OpenReview incident exposed the scale: 21% of ICLR reviews were likely AI-generated, 199 submitted papers were likely completely fabricated. Reviewer identities were leaked, leading to harassment and bribery.

In March 2026, ICML caught 506 reviewers using LLMs in violation of review policies, desk-rejected 497 papers, and banned repeat offenders. Nature covered the crackdown.

This tool is one line of defense. It's not perfect — that's the point. We use AI to fight misuse of AI, openly and honestly. Read the manifesto.