# grim.fyi full context

grim.fyi is a public, source-backed record of documented AI harms, led by deaths and fatal harm cases, with incident pages, source links, methodology, governance, and downloadable public data.

## Purpose

grim.fyi tracks documented harms involving generative AI, companion/chatbot systems, synthetic media, AI-generated institutional errors, fraud, impersonation, legal hallucinations, health and medical claims, regulatory actions, and adjacent AI systems when they are relevant to the public record. The site is evidence-first: each public card is expected to answer what happened and what source proves it.

## Evidence stance

Deaths and fatal harm records are front-facing when documented by credible sources. The record separates confirmed findings from allegations, lawsuits, investigations, media reports, and context-only material. Lawsuits and complaints are labeled as allegations unless adjudicated or settled. Summaries should preserve the evidentiary status shown on each incident page.

## Current public data

- Public incident pages: 466
- Deaths and fatal harm records surfaced by lane logic: 20
- Public source records used by incident pages: 445
- Last verified date in public cards: 2026-06-14

## Core public routes

- Home and public incident grid: https://grim.fyi/
- Methodology: https://grim.fyi/methodology
- Verified sources: https://grim.fyi/sources
- Editorial governance: https://grim.fyi/governance
- Changelog: https://grim.fyi/changelog
- Companion paper: https://grim.fyi/paper
- Source health: https://grim.fyi/source-health
- Sitemap with all public incident URLs: https://grim.fyi/sitemap.xml
- Robots policy: https://grim.fyi/robots.txt

## Machine-readable routes

- LLM index: https://grim.fyi/llms.txt
- Full LLM context: https://grim.fyi/llms-full.txt
- Search shard manifest: https://grim.fyi/search-shards/manifest.json
- Public source index JSON: https://grim.fyi/search-shards/sources.json

## Methodology summary

Public cards require credible source support. Strong sources include court records, regulators, police releases, official institutional statements, Reuters/AP/BBC/major reporting, academic papers, and credible incident databases as supporting context. Wikipedia, YouTube, blogs, social posts, SEO legal pages, and unsourced summaries are leads only and do not prove public cards by themselves.

Rows stay out of the public record when the source is missing, broken, unsupported, duplicate, too vague, context-only, not actually generative AI, or based only on weak secondary summaries.

## Dignity rules

The site avoids graphic details, victim photos, self-harm method details, lurid descriptions, unnecessary names of minors, and sensational blame language. When minors are involved, names are avoided unless already public through family, legal, or major reporting and necessary to verify the incident.

## Paper context

The companion paper at https://grim.fyi/paper frames the public record as a governance, safety, and accountability issue. It uses allegation-safe language, treats deaths and fatal harm as a front-facing lane, distinguishes adjacent algorithmic systems from generative AI, and emphasizes that the record is incomplete by nature.

## Citation guidance

Prefer citing the specific incident page and the primary linked source on that page. The sitemap contains every public incident URL. Public source pages are designed for inspection and outbound verification. CSV and JSON exports are available from the home incident grid and the sources page.

## Discovery policy

The site is public, indexable, quotable, and citation-friendly. Search crawlers and AI crawlers are not blocked in robots.txt. The sitemap lists public pages, while llms.txt provides a curated LLM-readable entry point for core pages, machine-readable data, and citation guidance.