Database

AVID is a database of failure modes in general-purpose AI (GPAI) systems, including LLMs, API-only AI systems, developer tooling, and end-to-end applications and agents. The two base data classes are Report and Vulnerability. A report captures one concrete occurrence with supporting evidence; a vulnerability (vuln) captures a recurring failure mode.

Records in this database can be mapped to multiple taxonomy and risk frameworks, including the AVID taxonomy, CVSS risk scores, and MITRE ATLAS. This page lists the current reports and vulns in our database. To learn more about the database and usage, refer to our documentation.

Note: Older AVID records (before 2025) were curated under a broader AI/ML scope; these should be interpreted as legacy entries relative to the current GPAI-focused scope.

The definition of an “AI vulnerability” is still evolving across the ecosystem, so AVID currently uses a working definition. In this release, we are prioritizing report-level evidence and have not yet published new vulnerability records.

Reports

Reports are occurrences of a GPAI failure mode. We classify reports in four types, in increasing degree of quantitative evidence:

1. Issue: qualitative evaluation based on a single sample or handful of samples,
2. Advisory: qualitative evaluation based on multiple Incidents,
3. Measurement: quantitative evaluation with associated data and metric,
4. Detection: A Measurement deemed critical by a threshold or statistical test.

List of Reports

2026

2025

2023

2022

Vulnerability

Vulnerabilities can be linked to one or more taxonomies through tags. In AVID taxonomy, these tags denote the risk domains (Security, Ethics, Performance), related (sub)categories, and lifecycle stages. A vulnerability in AVID can pertain to one or more of the three levels: dataset, model, or system.

List of Vulnerabilities

2023

2022