How RedEval.ai Works

Everything you need to know about adversarial testing for AI systems — from attack techniques to how results map to your governance frameworks.

DESCRIBE

Tell us what your AI does and what data or tools it can access

GENERATE

Attack batteries and multi-turn scenarios are built for your risk categories

EVALUATE

Dual-model judging scores every response for safety and relevance

GETTING STARTED

What is RedEval.ai?

Who is it for?

Is it self-serve or managed?

How do I get access?

WHAT WE TEST

What attack techniques does RedEval use?

What are multi-turn (crescendo) attacks?

Can you test tool-using or agentic systems?

What about indirect and RAG injection?

RISK CATEGORIES

What is Prompt Injection?

What is PII Extraction?

What is Role Confusion?

What is Harmful Content?

What is Bias & Fairness testing?

UNDERSTANDING RESULTS

What do the severity levels mean?

How are responses judged?

What is a Passed vs Failed result?

FRAMEWORKS & COMPLIANCE

Which frameworks does RedEval align with?

Can I re-test after a fix?

Ready to put your AI to the test?