Welcome! aiXamine is a SaaS service for evaluating LLMs and their data against various safety issue and security threats. Our goal is to provide individuals and organizations with a reliable service to examine the safety and security of their models before they are integrated into end-user applications. Below, you can find detailed information on how to use aiXamine and learn more about the supported examination services.

📄 Our paper is now available on arXiv!

How To

Search Examination Reports

Create an Examination

Compare Models in a Leaderboard

View an Examination Report

Manage User Account

Learn About

Adversarial Robustness

Code Security

Fairness & Bias

Hallucination

Jailbreaking

Model & Data Privacy

OOD Robustness

Over Refusal

Safety & Alignment

Disclaimers

1. Datasets used in aiXamine contain dialogue that may be considered offensive or harmful.
2. During examinations, we allow up to 1% of model responses to be empty. Scores are calculated only on the remaining valid responses. If more than 1% of responses are empty, the evaluation is considered failed.

References

References