The writeup on Git Hub reports 3,607 user‑reported incidents of AI agents misbehaving, collected from Git Hub issues, Hacker News, Less Wrong, and X. These reports are normalized into a shared record format and labeled by an LLM classifier across fourteen misbehavior categories.

Incidents are multi‑label, so category counts exceed the total. Of the reported cases, 40.7% (1,468) are negligible, 38.1% (1,373) are minor, 17.1% (618) are significant, and 3.4% (121) are severe, with 0.7% (27) unrated.

The monthly severity mix shows the share of each harm level from Jan 2025 through Jun 2026, with earlier months omitted due to small samples. The methodology notes that only the published subset (confidence ≥ 0.9) is shown; X posts and AI Incident Database records are gathered but not republished because of licensing constraints.

All collection and classification code, plus the full pipeline, are openly available on Git Hub. The data reveal that while most incidents cause little or recoverable harm, a non‑trivial fraction leads to real cost or irreversible damage.