As artificial intelligence embeds itself deeper into enterprise infrastructure, a critical question has gone largely unanswered: How secure is the AI model your organization actually relies on? F5 Labs is now stepping in with a definitive answer — and the results may surprise security teams worldwide.
F5 announced two new industry benchmarks — the Comprehensive AI Security Index (CASI) and the Agentic Resistance Score (ARS) — designed to give organizations standardized, monthly-updated metrics for evaluating the real-world security performance of AI models before and after deployment.
What Makes These Rankings Different
Built on F5’s acquisition of CalypsoAI, the leaderboards draw from one of the largest AI vulnerability libraries in existence, incorporating over 10,000 new attacks prompts every month and more than a year of accumulated attack data.
CASI goes beyond a simple pass/fail. It measures average model performance under normal conditions, evaluates the trade-off between safety and capability through a Risk-to-Performance Ratio, and even calculates the Cost of Security—the financial cost of securing a model relative to its risk score.
ARS takes the analysis further by simulating prolonged, adaptive AI-agent attacks — not one-off prompt injections, but sustained campaigns that reason, adapt, and probe for weaknesses over time. It evaluates three dimensions: how sophisticated an attack must be to succeed, how long defenses hold under pressure, and whether failed attacks inadvertently leak exploitable signals.
Why This Matters Now
“Deploying unverified AI models into critical infrastructure is not innovation; it is negligence,” said Kunal Anand, Chief Product Officer at F5.
The stakes are real. AI systems are increasingly managing APIs, sensitive data pipelines, and automated decision-making — all prime targets. Without standardized security benchmarks, enterprises have been essentially flying blind.
The Bigger Picture
The leaderboards complement F5’s recently launched AI Guardrails and AI Red Team tools, forming a comprehensive security ecosystem. Monthly companion research from F5 Labs will explain notable score shifts and provide deep dives on emerging attack vectors.
For security teams navigating a rapidly evolving threat landscape, these rankings may become as essential as any firewall.
