F5 XC EMEA Workshop > Class 6 - F5 AI Security > F5 AI Red Team Source |
F5 CASI Leaderboard¶
This is the last part of our lab. We hope you enjoyed it and would like to mention one last topic: the F5 CASI Leaderboard.
The CASI Leaderboard (Comprehensive AI Security Index) is a security-focused ranking of large language models published by F5 Labs. It is designed to help organizations evaluate and compare AI models based not only on performance, but on their resistance to security threats such as prompt injection, jailbreaking, and adversarial misuse.
CASI emphasizes that AI adoption should consider security posture alongside capability, particularly for enterprise and regulated environments.
You can find the F5 CASI Leaderboard at https://www.f5.com/labs/casi.
A detailed explanation of each metric is available at https://www.f5.com/labs/articles/introducing-the-casi-leaderboards
What CASI Measures¶
The CASI Leaderboard evaluates models across several standardized metrics:
CASI Score¶
The primary metric of the leaderboard. It reflects how resistant a model is to known and emerging attack techniques based on extensive red-team testing.
Performance¶
A measure of the model’s general task capability and effectiveness. This ensures security is evaluated in context, rather than in isolation.
Risk-to-Performance Ratio (RTP)¶
A comparative metric that balances security risk against model capability. It helps identify models that offer strong performance without disproportionately high security risk.
Cost of Security (CoS)¶
An estimate of the operational or architectural cost required to achieve an acceptable security level for a given model, relative to its performance.
How the Leaderboard Is Used¶
The CASI Leaderboard typically presents:
- A ranked list of leading AI models
- Comparative scores across security and performance metrics
- Regular updates as new models and attack methods emerge
The rankings are derived from continuous adversarial testing conducted by F5’s AI security research team.
Why the CASI Leaderboard Matters¶
As AI models become more widely deployed in production systems, security risks increase. The CASI Leaderboard helps organizations:
- Compare AI models using security-first criteria
- Understand trade-offs between performance, risk, and cost
- Make more informed decisions for enterprise and high-risk deployments
A higher CASI score indicates stronger resistance to misuse and exploitation, making the leaderboard particularly relevant for security-conscious adopters.
Summary¶
The CASI Leaderboard is an AI security benchmarking framework that shifts the focus from raw performance alone to secure, responsible AI deployment. It provides a structured, transparent way to assess how well AI models stand up to real-world adversarial threats.