Cisco AI Defense vs Protect AI

Cisco AI Defense is Cisco's security product for enterprise AI applications, models and agents. Cisco unveiled it on January 15, 2025 and said it would be available that March. It is built on Robust Intelligence, the AI security startup Cisco announced it would acquire on August 26, 2024 and finished acquiring on September 24, 2024; Dark Reading called Robust Intelligence the core component of AI Defense. The product is sold by Cisco Systems, Inc., which was incorporated in 1984 and is based in San Jose; this profile lists Cisco's founding year and headquarters, as the directory does for other big-company products. Cisco's May 2026 data sheet splits the product into four parts. AI Cloud Visibility discovers AI models, applications, connected data sources, MCP servers and agents, including assets in Amazon Bedrock, Google Vertex and Azure AI Foundry. AI Supply Chain Risk Management scans model files, repositories and MCP servers and tools, and gives each asset a risk score. AI Model and Application Validation runs algorithmic red teaming with more than 200 attack techniques and threat subcategories. AI Runtime Protection applies safety, security and privacy guardrails, enforces MCP policies at runtime and protects agent behavior. Runtime checks can run at the network level through Cisco Multicloud Defense, through a gateway, or through an Inspection API. Hybrid deployments (AWS EKS, Azure AKS, GCP and on-premises AI PODs on OpenShift) keep sensitive data in the customer's environment. A separate AI Access module, for employee use of third-party AI apps, comes through Cisco Secure Access and needs its Advantage package. **Agent and MCP security** In February 2026 Cisco added an AI bill of materials, an MCP Catalog, multi-turn red teaming of models and agents, real-time agentic guardrails and an integration with NVIDIA NeMo Guardrails; the release gave no availability dates. In March 2026 it launched AI Defense: Explorer Edition, a self-service red-teaming tool Cisco offers at no upfront cost, and announced an Agent Runtime SDK for Amazon Bedrock AgentCore, Google Vertex Agent Builder, Azure AI Foundry and LangChain, plus DefenseClaw, an open-source agent security framework. **Frameworks** Cisco says AI Defense has mapped to NIST, OWASP and MITRE frameworks since launch. Its data sheet names the NIST AI RMF, MITRE ATLAS and the OWASP Top 10 for LLMs, and red-team reports also map to Cisco's own Integrated AI Security and Safety Framework. We found no documented support for the EU AI Act or ISO 42001. **Who it's for** Large enterprises that already use Cisco security products (Secure Access, Multicloud Defense, Splunk) and want AI discovery, red teaming and runtime guardrails from one vendor. Cisco doesn't publish prices.

Protect AI was a Seattle AI security company founded in 2022 by Ian Swanson (CEO), Daryan Dehghanpisheh and Badar Ahmed, who had previously worked at AWS, Oracle and DataScience.com. It raised $108.5 million: a $13.5 million seed round co-led by Acrew Capital and Boldstart Ventures (December 2022), a $35 million Series A led by Evolution Equity Partners (July 2023) and a $60 million Series B led by Evolution Equity Partners (August 2024), with Salesforce Ventures and Samsung among the investors. GeekWire reported 75 employees and offices in Seattle, Berlin and Bangalore in August 2024. Protect AI bought huntr, Rebuff, Laiyer AI and SydeLabs. Palo Alto Networks announced its intent to acquire Protect AI on April 28, 2025, the same day it launched Prisma AIRS, and completed the deal on July 22, 2025. It did not disclose the price; SiliconANGLE cited GeekWire sources putting it above $500 million. Ian Swanson joined Palo Alto as VP, Product, Prisma AIRS, and Prisma AIRS 2.0 (October 28, 2025) "completes the native integration" of Protect AI. Protect AI is no longer sold separately: protectai.com now redirects to the Prisma AIRS page, although some old blog posts are still online. Buyers should evaluate Prisma AIRS. **What it sold** Protect AI sold AI security, not governance. Guardian, built on its open-source ModelScan tool, scanned machine learning models for unsafe or malicious code and enforced policies on which models could be used. In October 2024 Hugging Face added the Guardian scanner to its Hub, and by April 2025 Protect AI had scanned 4.47 million model versions there. Recon tested AI applications before deployment with a library of 450 simulated attacks, and Layer filtered malicious prompts in production. Its earlier AI Radar product built a machine learning bill of materials. The huntr community, with more than 15,000 members in August 2024, reported AI and ML vulnerabilities. Palo Alto says the deal brought model scanning, posture management, AI red teaming, runtime protection and AI agent security to Prisma AIRS, but it doesn't map each Protect AI product to a Prisma AIRS module. **Open source** LLM Guard, Protect AI's open-source library for scanning prompts and responses, is archived: its README says the project and its Hugging Face models are no longer developed or maintained. **Frameworks** The Protect AI pages still online don't document a mapping to an AI framework, so this profile lists none. Prisma AIRS maps its AI Red Teaming attacks to the OWASP Top 10 for LLMs, MITRE ATLAS and NIST.