AI Risk Assessment 101

This guide is for new users with no prior experience in AI testing. You'll learn how to evaluate AI behavior, spot risks, and submit clear feedback.

What is AIRTA Systems?

AIRTA Systems is an AI Risk Management Platform where compliance is the product. Through The Consolidated AI Product Safety Pipeline, it integrates agentic AI risk assessment, compliance documentation, post-market monitoring, and safety certification. Companies run programs and invite teams to test their AI systems. As a tester, you join or create a team; when your team is invited to a program, you'll see it on the platform and can test within scope.

What is AI Risk Assessment?

AI risk assessment checks whether an AI is safe, fair, and works as expected. You'll test AI applications, models, and prompts for real-world issues-prompt injection, hallucinations, bias, security vulnerabilities, misuse, and consumer harms-so companies can fix problems before they reach users.

---

Step 1: Create a tester account

Go to the arrange a demo page to get started and create your free account. Select the tester (AI Safety Researcher) role.

Step 2: Complete your profile and submit for vetting

From your dashboard, complete your profile and submit for vetting. Approval is required before you can participate in programs.

Step 3: Join or create a team and get invited to programs

Companies invite teams to test their AI systems, not individuals. Join an existing team or create one. Once your team is invited to a program, you'll see it on the AI Safety Testing Programs page and can start testing within the program's scope.

Step 4: Pick what to test (within scope)

Programs list the apps and features in scope. Typical areas:

• AI applications & chatbots: Incorrect or harmful answers, prompt injection, data leakage.
• AI models: Hallucinations, bias, reliability, and robustness under edge cases.
• Prompts & agents: How the AI responds to different inputs, misuse, or safety failures.

Always stay within the program's scope and use the AI Safety Testing Programs page to access programs your team has been invited to.

---

Step 5: Look for problems

Try different inputs and scenarios. Look for:

• Incorrect, harmful, or misleading outputs (hallucinations, misinformation).
• Unfair or discriminatory behavior (bias).
• Ways to trigger bad behavior (prompt injection, misuse).
• Security issues (data leakage, API misuse).
• Failures in validity, reliability, or robustness.

---

Step 6: Write down what you find

Use a clear structure so your report can be validated quickly:

1. Scenario: What you tried (e.g. "I asked the AI about [topic]" or "I used [feature] with [input]").
2. Expectation: What should have happened.
3. Observation: What actually happened-quote the AI's response or describe the behavior.
4. Risk: Why it matters (e.g. harm, compliance, safety).
5. Next step: Smallest fix you'd suggest, if any.

See Write a Clear AI Safety Report for more detail.

---

Step 7: Submit your findings

Submit your report through the AIRTA Systems platform for the program you're testing. Include screenshots or exact text of the AI's response, and stick to the program's scope and rules (see Don't Cause Harm for safe testing).

Reputation points & compensation

• Gain reputation points for each validated submission (e.g. 10 to 5000 points based on severity).
• Compensation is defined per program and outlined in the project brief.
• Higher reputation helps your team get invited to more and premium programs.

---

Tips for beginners

• Start with simple tests and stay in scope.
• Be ethical and follow safe harbour and program rules.
• Write clear, structured reports so findings can be validated.
• Focus on consumer safety and real-world impact.

---

What's next?

Once you're comfortable with basic testing:

• Explore more techniques in the Academy.
• Build reputation through validated findings so your team gets invited to more programs.
• Read How to Build Your Reputation as an AI Safety Researcher.
• Understand Where You Fit: Human-in-the-Loop Testing in the AIRTA Systems structure.