F5 AI Guardrails vs Robust Intelligence

F5 AI Guardrails is F5's runtime security product for AI models, applications and agents. F5 sells it alongside F5 AI Red Team, its automated adversarial testing product, and presents the two as a pair: findings from Red Team can be turned into guardrail policies. Both come from CalypsoAI, the AI security company F5 announced it would acquire on September 11, 2025 for $180 million in purchase consideration, mostly cash. The deal closed on September 26, 2025. F5 introduced both products when it announced the deal had closed, and made them generally available in January 2026 as part of its Application Delivery and Security Platform. F5, Inc. was incorporated in 1996 and is based in Seattle; this profile lists the company's founding year and headquarters, as the directory does for other big-company products. F5 AI Guardrails blocks prompt injection and jailbreaks, detects and prevents leakage of personal data and custom data categories, filters toxic or harmful output, and audits or blocks unauthorized tool calls and actions by agents. Policies can be written in natural language. Every enforcement action is logged with attribution, and logs can be exported to third-party SIEMs. It works with any model and runs in AWS, Azure or Google Cloud, in private clouds, or on premises, including fully air-gapped environments. F5 AI Red Team sends a swarm of attacker agents to run static, prompt-based and multi-turn attacks, including indirect prompt injection and data-leakage tests on RAG applications. F5 says its attack library grows by more than 10,000 patterns a month. Red Team runs in the major public clouds or on premises as a certified Red Hat OpenShift operator. A feature called F5 AI Remediate drafts guardrails from Red Team findings, and a person approves each one before it goes live. **Gateway and platform** On June 22, 2026 F5 launched the F5 AI Security Platform, which groups AI governance, AI discovery, testing with AI Red Team and runtime protection with AI Guardrails, plus an audit trail across AI interactions. At the same time it announced the acquisition of SurePath AI, which adds network-based discovery of AI use, including shadow AI. F5's AI Gateway now combines a model gateway, an MCP gateway (with an MCP server registry and per-tool allow and deny controls) and AI Guardrails under one control plane. **Compliance coverage** F5 says Guardrails has automated auditing templates for GDPR, HIPAA and the EU AI Act, out-of-the-box policy controls aligned to PCI, PHI and the EU AI Act's prohibited-risk categories, and audit-ready logs. It also says Red Team findings can support compliance programs aligned with the NIST AI RMF and the EU AI Act, and it publishes an infographic aligning its products with the OWASP Top 10 for LLMs. We found no article-by-article mapping to the EU AI Act. **Who it's for** Large enterprises, including regulated financial services and healthcare organizations, that need AI guardrails they can host themselves or run air-gapped. F5 says both products are deployed at Fortune 500 companies. It also suits F5 customers who want AI controls on the same platform as their application delivery and API security.

Robust Intelligence was an AI security company founded in 2019 by Yaron Singer, a Harvard computer science and applied mathematics professor who became its CEO, and Kojin Oshiba, his former student. It was based in San Francisco and also had operations in Israel. Sequoia Capital led its seed and Series A rounds ($14 million combined, announced in October 2020), and Tiger Global and Sequoia led a $30 million Series B in December 2021, bringing its funding to about $44 million. Cisco Investments was also an investor. Cisco announced its intent to acquire Robust Intelligence on August 26, 2024 and completed the deal on September 24, 2024. The price was not disclosed. Robust Intelligence is no longer a separate product: its technology is the core of Cisco AI Defense, which Cisco unveiled on January 15, 2025 and said would be available that March. Buyers should evaluate Cisco AI Defense. **What it sold** Robust Intelligence started with RIME (the Robust Intelligence Model Engine), built around an "AI Firewall" that wraps a model, checks its inputs and outputs, and stress-tests it. Its later platform had two parts. AI Validation automatically tested models for security and safety vulnerabilities (such as prompt injection and data poisoning) before and after deployment, scanned open-source models, data and files for supply chain vulnerabilities such as malicious pickle files, recommended guardrails and produced reports. AI Protection applied those guardrails in production. It ran as SaaS in Robust Intelligence's private AWS VPC or as a hybrid deployment with a self-hosted data plane, and it integrated with observability and SIEM tools such as Splunk, Datadog, CrowdStrike and AppDynamics. The company also co-developed the AI Risk Database and said it co-authored NIST's adversarial machine learning taxonomy and contributed to the OWASP Top 10 for LLM Applications. **Frameworks** Its product page says tests were mapped to the OWASP Top 10 for LLM Applications, MITRE ATLAS, NIST's adversarial machine learning taxonomy, the EU AI Act and the White House executive order on AI. It does not mention the NIST AI RMF or ISO 42001. Cisco AI Defense maps to the NIST AI RMF, MITRE ATLAS and the OWASP Top 10 for LLMs. **Status** The robustintelligence.com site is still online and describes AI Validation and AI Protection without mentioning Cisco. We found no current way to buy Robust Intelligence separately from Cisco.