F5 has introduced enhanced threat intelligence resources enabling enterprise security leaders to reliably measure and compare the risk profiles of all popular AI models. From the F5 Labs threat research team, Comprehensive AI Security Index (CASI) and Agentic Resistance Score (ARS) leaderboards provide standardized, monthly-updated benchmarks backed by leading research, pairing real-time attack intelligence with expert analysis of evolving AI attack vectors.
Stemming from F5’s acquisition of CalypsoAI, these industry-leading security resources include one of the largest AI vulnerability libraries, uniquely updated with more than 10,000 new attack prompts each month, and utilizing over one year of accumulated attack data. F5 Labs now offers enterprises and the wider security community a powerful and consistent means to evaluate, compare, and select AI models and providers prior to production, based on their ability to address key challenges and real-world application security threats.
“Deploying unverified AI models into critical infrastructure is not innovation; it is negligence. Organizations need a way to continuously quantify resilience. F5 Labs AI Leaderboards offer that standard. These rankings isolate specific weaknesses in the model layer, giving security teams the intelligence they need to govern inference and block attacks before they happen,” said Kunal Anand, Chief Product Officer at F5.
With enhanced visibility, leaderboards from F5 Labs help security teams holistically identify and shore up vulnerabilities to improve defenses alongside the F5 Application Delivery and Security Platform, from safeguarding APIs and data to preventing DDoS attacks to accelerating DevSecOps through automation and scalability.
Assessing today’s AI model landscape
The rapid integration of AI into every facet of business operations has brought with it a requirement for the robust security validation of an organization’s chosen models. Designed to help security practitioners answer, “How secure is my model?”, F5 Labs’ leaderboards establish metrics around the paths of least resistance and minimum compute resources required to complete both simple and complex attacks.
Along with a straightforward ranking, F5 Labs’ Comprehensive AI Security Index (CASI) offers metrics such as:
- Average Performance: Baseline model performance measured across standardized tasks under normal operating conditions
- Risk-to-Performance Ratio: Insight into the tradeoff between model safety and performance
- Cost of Security: The current inference cost relative to the model’s CASI, assessing the financial impact of security
Supplementing CASI, F5 Labs’ Agentic Resistance Score (ARS) evaluates how AI systems withstand sustained, adaptive attacks by an AI agent tasked with achieving a goal. Rather than attempting to execute one individual prompt, AI agents conduct prolonged interactions with models, applying reasoning and psychological methods in an attempt to bypass security guardrails. ARS assesses AI systems across three core dimensions:
- Required Sophistication: The minimum level of attacker ingenuity needed to successfully compromise the AI system
- Defensive Endurance: How long the system remains secure under prolonged, adaptive, multi-step attacks
- Counter-Intelligence: Whether failed attacks inadvertently expose signals or system behavior that could enable future exploits
