Introduction to the database Vulnerabilities and Exposures AI models

Why?

AI is broken, and we are here to fix it

The purpose of the CVE-AI database is to collect and systematize reports on vulnerabilities and weaknesses of openly available AI models.

We are cataloguing vulnerabilities for at least three purposes:

• it will allow model developers to better understand what kind of adversarial model analysis they need to take into account, and they will see the effects of this analysis on other models,
• it will allow model users to figure out what kind of problems they can expect if they use a specific model
• it will allow people working on adversarial model analysis to better understand what techniques work and what don’t, and as a further consequence allow to develop mechanisms to make models immune to important classes of attacks.

How?

>>> Here you will find a list of reports for the vulnerability database.

>>> Here you will find instructions on how to submit your own report.

The vulnerability database is being developed in parallel to the incident database of AI models, such as AI Incident Database or OECD AI Incidents Monitor.

the CVE-AI database deals with reports related to adversarial analysis of models, while the incident database describes accidental, unplanned misuses of models.