F5 said it has introduced new threat-intelligence benchmarks to help security teams compare the security risk of widely used AI models. The resources include two leaderboards from F5 Labs that the company says are updated monthly: a Comprehensive AI Security Index (CASI) and an Agentic Resistance Score (ARS), intended to provide standardized measures of model risk based on attack intelligence and analysis of AI attack methods.

F5 said the benchmarks are supported by an AI vulnerability library that it claims is updated with more than 10,000 new attack prompts each month, using accumulated attack data. The intent is to help organizations evaluate and compare models and providers before deploying them in production environments.

CASI is described as a baseline model-security benchmark that also reports supporting metrics such as baseline performance under standard tasks, a risk-to-performance tradeoff measure, and an estimate of inference cost relative to the model’s security score. ARS is described as a measure of how well models withstand sustained, adaptive attacks carried out by an AI agent over multiple steps, focusing on the level of attacker sophistication required, how long defenses hold under prolonged attempts, and whether failed attempts leak useful signals that could enable future attacks.

F5 also said these leaderboards relate to its AI security testing and guardrail capabilities and that it plans to publish recurring companion analysis explaining score changes and notable developments in AI security.

Join the community of 2M+ industry professionals.

Subscribe to Newsletter to get latest insights & analysis in your inbox.

All about ETCISO industry right on your smartphone!