Viewing the results

Although we haven’t run the attack ourselves, we will look at the results of an attack that has already run.

1. In the main left tab go to Reports, click on the Full report report name.

   3 different models have been tested: gpt-oss-20b-deepinf, qwen3-32B-deepinfra, llama-33-70B-instruct-turbo-deepinfra

   First observe the CASI score for each of the tested models. This score is a composite metric designed to measure the overall security of a model and it is based on the Signature attacks. The higher the number, the more secure the model is.

   Scrolling down, you will see the vulnerabilities that have been found and the respective recommendations.

2. Observe the Agentic Warfare score. This score, also called ARS ( Agentic Resistance Score ), represents a quantitative measure of an AI system’s defensive strength, rated on a scale of 0 to 100. A higher ARS indicates that a system requires a more sophisticated, persistent, and informed attacker to compromise it.

   Now click on View agentic fingerprints

   You will see the agentic attack that has been performed with multiple methods.

   While this attack has not been succesfull you can clearly see the attack logic and paths taken.

   Click on the drop-down under Attacks and compare the approach of each attack type.

3. In the main left tab go to Reports, click on the Full on Attack Signatures report name.

   Now if you click on View raw data you will get a list of all attacks that have been performed, the prompt, and each individual result.