The dataset

Artifacts

• JSON /api/failures.json: every live entry, all data fields, source URLs and types, taxonomy version, generated date.
• CSV /data/aifi-latest.csv: the stable alias; a dated snapshot (aifi-snapshot-YYYY-MM-DD.csv) ships alongside it at every build.
• RSS /rss.xml: the 50 newest entries by index date, GUIDs are the FI ids.
• Taxonomy /data/taxonomy.json: the versioned classification standard (currently v1.0.0).

Field dictionary

• id Permanent entry ID (FI-0001 style); stable across edits and the RSS GUID
• slug / url Permanent entry URL; merged entries redirect to their canonical record
• headline The entry headline
• company Named company, or null when the entry is anonymized
• is_anonymized True for steward-verified entries published under NDA
• industry One of the taxonomy industries
• failure_mode The mechanism classification (8 modes)
• ai_surface Where the system shipped (12 surfaces)
• system_class generative or predictive
• severity Catastrophic, High, Medium, or Low, per the /about rubric
• date_of_incident / date_indexed When it happened and when it entered the index
• public_visibility / regulatory_exposure / customer_impact / financial_impact / time_to_disclosure The operational-cost fields
• sources Source title, URL, type, and date; sources_count and source_urls in the CSV

License

The data fields (everything in the JSON and CSV artifacts) are licensed CC-BY 4.0: attribution required, reuse welcome. The narrative prose on entry pages (summaries, analysis, the steward note) carries standard copyright; quoting with citation is permitted and encouraged. Cite as "AI Failure Index, stewarded by Realm Labs" with the entry URL.

Update cadence

Entries are added as verified and the artifacts regenerate at every publish. Last indexed: June 16, 2026. 532 entries, 1379 sources cited. The quarterly correction sweep and any merges are logged in the version history on the methodology page.