Show HN: Marcus – catches the fake source links AI search engines invent
Category: ai-ml
Tags: ai-auditing, fact-checking, citations, open-data
Score: 5.7/10 (Innovation: 6, Technical: 5, Documentation: 5, Utility: 6)
Marcus is open-source software that audits AI search engines by checking whether their cited sources are real and correct. It ran a public experiment with 800 citations from four engines, revealing that 69.5% of answers were wrong and 51% of links were fabricated, dead, or misattributed. The project provides a dataset, grader, and audit records to make AI agents prove their work.
Target audience: AI researchers, developers building AI search, and fact-checking teams
Repository: https://marcus-ai.dev/on-the-record/
View on Hacker News