AVISE Documentations
AVISE (AI Vulnerability Identification & Security Evaluation) is a modular framework for developing automated Security Evaluation Tests (SETs) to identify vulnerabilities in and assess the security of AI systems. AVISE provides a scalable, automated approach to red teaming AI systems. It allows developers and researchers to deploy consistent, rigorous security evaluation tests across a wide range of different types of AI systems from Large Language Models to specialized Continual Learning systems.
The core element of AVISE is its extensible acrhitecture. The modular design of the framework allows automated security evaluation tests to be developed for various AI types with the same framework, without reinventing the wheel every time. And as novel adversarial techniques and AI types emerge, new BaseSETModules can easily be developed with the framework that allow development and deployment of new kinds of Security Evaluation Tests, addressing the emerging needs.
To get started, you can browse our SET Registry to select pre-built tests that match your specific AI stack. If you are dealing with a unique edge case, or wish to identify vulnerabilities that we haven’t yet developed automated Security Evaluation Tests for, you can chat with us in Discord. We are always eager to hear user and developer feedback on how we could improve AVISE. Additionally, extending AVISE with new SETs is straightforward - we suggest to get familiar with BaseSETModules, SETs, and Contributing within these documentations, if you’d like to develop your own Security Evaluation Tests. For technical support, you can also hop in to Discord and we’re happy to help you.
Note
This project is under active development. For any questions, contributions, or improvement suggestions, the best way to reach us is at the project Discord server.
Reference:
Extending: