F5 completes acquisition of CalypsoAI, introduces F5 AI Guardrails and F5 AI Red Team

Today, F5 is pleased to announce that we have completed our acquisition of CalypsoAI, a pioneer at the frontier of AI security, and we are introducing the latest addition to securing our digital roads: F5 AI Guardrails—comprehensive runtime security for AI models and agents. With this new addition, F5 can now provide a model-agnostic solution that secures AI data, combats adversarial threats, and ensures responsible AI governance across all interactions. The AI landscape is constantly evolving with millions of proprietary and open-source models, as well as their fine-tuned variations. The present and future of AI security is not limited to a few monolithic LLMs, but rather a cascade of proprietary and public AI innovations, each requiring consistent privacy and security, independent from any single model provider. Every model carries a different portfolio of risks, and thus, every model needs tailored Guardrails to keep AI interactions from drifting beyond acceptable boundaries.

You can’t secure what you can’t see. Across all physical and digital highways, there are countless points where guardrails could be useful. While known accident locations or high-traffic intersections are naturally prioritized, a more strategic, preventative approach is needed. F5 AI Red Team acts as the eye in the sky, analyzing the landscape to simulate the real-world chaos of rush hour traffic, the actions of worst-case road rage drivers, and the vulnerabilities they could expose. By identifying not only where accidents in AI systems are likely but where they would have the greatest impact, AI Red Team informs exactly where guardrails need to go and how urgently they should be placed.

We are introducing AI Red Team to deliver this capability in stride, unleashing swarms of autonomous agents to simulate thousands of attack patterns and hunt vulnerabilities in AI systems. These agent swarms accelerate discovery of both common AI vulnerabilities as well as the most obscure, worst-case exploits that security teams lack the time to consider. What’s more, simulated attack techniques are rooted in our preeminent AI vulnerability database, a vast library updated regularly with over 10,000 new attack patterns each month to reflect emerging trends. The insights are valuable on their own, but with a few clicks, teams can translate every insight into active AI Guardrails.

The ‘what’ and ‘where’ of AI threats are foundational requirements to maintaining security posture but understanding ‘how’ and ‘why’ is essential for improving posture. The opaque ‘black box’ nature of AI systems has historically made this difficult, calling for an entirely new approach. Assume your AI hits a Guardrail and a prompt gets blocked. Under traditional methods like static keyword or regex filtering, you would know the user, datetime, and trigger, but you wouldn’t have the full picture of what occurred. Crucially, F5 AI Guardrails and AI Red Team prioritize explainability, helping customers understand why an AI system made a particular decision or took a particular action.

With advanced capabilities like Agentic Fingerprints and Outcome Analysis, teams now have granular insight into every AI interaction, with detailed reasoning as to why a prompt was accepted or blocked, as well as the context leading up to that decision. This level of granularity is not only critical for informing security teams, but it radically simplifies compliance with accessible audit trails and ready-made templates by regulatory framework.