A structured, anonymized catalog of real-world data security incidents in GenAI deployments.
Status: Accepting contributions — this dataset is being built from scratch with the community.
Documented incidents where GenAI systems experienced data security failures in production or staging environments. This includes data leakage events, cross-tenant bleed, RAG exfiltration, agent credential exposure, training data extraction, vector store compromise, telemetry over-collection, and any other incident aligned with the DSGAI risk taxonomy (DSGAI01–DSGAI21).
This dataset is modeled on existing incident databases (e.g., AIAAIC, AI Incident Database) but focused specifically on data security failure modes in GenAI systems.
Contributions should include where possible:
- Incident ID — Unique identifier (assigned on merge)
- Date — When the incident occurred or was disclosed
- Summary — Brief description of what happened
- DSGAI mapping — Primary DSGAI entry (DSGAI01–DSGAI21)
- Attack vector — How the data security failure occurred
- Data types affected — PII, PHI, credentials, IP, model weights, embeddings, etc.
- Root cause — Misconfiguration, missing control, architectural flaw, insider, etc.
- Impact — Scope of data exposed, regulatory consequences, operational disruption
- Detection method — How the incident was discovered
- Mitigations applied — What was done to remediate
- Source — Public advisory, news report, research paper, or anonymized self-report
- Industry / sector — If known and non-identifying
All submissions must be anonymized:
- No organization names, employee names, or customer names
- No IP addresses, internal hostnames, or system identifiers
- No proprietary model names unless publicly known
- Generalize details that could identify a specific organization (e.g., "a financial services company" not "Bank X")
Self-reported incidents from your own organization are welcome and encouraged — anonymization makes this safe.
Add entries as individual JSON or Markdown files and submit a pull request. See the main datasets README for general contribution guidelines.