AI Defense, part of Cisco's Secure Access platform, offers a unified security approach that includes various services such as zero trust network access and VPN-as-a-Service, and will be generally available to enterprises starting in March 2025.

As enterprises are expected to create hundreds to thousands of AI applications, Cisco recognizes the impracticality of implementing separate security measures for each application.

This innovative cloud security technology, introduced on January 15, 2025, aims to help developers identify and mitigate vulnerabilities in AI applications.

AI Defense employs detection models from Scale AI and integrates threat intelligence from Cisco's Talos and Splunk to automatically recommend safety measures for AI models.

Jeetu Patel, Cisco's Executive Vice President, stressed the necessity of balancing safety with speed in AI development to maintain competitiveness in the market.

The technology utilizes the Robust Intelligence platform, which Cisco acquired, to perform algorithmic red teaming that identifies vulnerabilities in AI models.

Cisco has unveiled a new security solution named Cisco AI Defense, designed to assist businesses in securely building, deploying, and managing AI applications.

AI Defense features self-optimizing capabilities that evolve over time and seamlessly integrate with existing data flows, thereby enhancing overall security measures.

The solution enables continuous validation of AI models, ensuring that applications perform as intended despite ongoing changes in data.

This initiative responds to security concerns highlighted in Cisco's 2024 AI Readiness Survey, where 71% of organizations expressed feeling unprepared to prevent unauthorized AI tampering.

Cisco claims that it can validate AI models in approximately 30 seconds, a remarkable improvement compared to the typical seven to ten weeks required by most organizations.

The increasing frequency of attacks on vulnerable AI models, including incidents of LLMjacking and cloud credential theft, highlights the critical need for robust AI security.