Cisco introduced a new product called Cisco AI Defense in early access this week, aiming to enhance AI cybersecurity in enterprise SecOps and to protect against malicious usage on corporate networks. The product is set to be released in March and is designed to address the gaps in AI cybersecurity as more mainstream enterprises deploy generative AI applications. Cisco AI Defense will analyze web traffic to identify the AI applications and services being utilized within a corporate network and determine if their use is malicious. It will have the capability to detect and block malicious prompts and outputs from large language models, similar to other AI gateway products recently announced. One of the key distinguishing factors claimed by Cisco is its automatic, algorithmic security validation for large language models as they are optimized within an enterprise, along with the automatic creation of suitable security guardrails.
During a keynote presentation at the Cisco AI Summit in New York, Jeetu Patel, the executive vice president and chief product officer at Cisco, emphasized the importance of validating models before putting them into production to ensure they function as expected and implementing the necessary guardrails. He highlighted the lack of a central vulnerability database for AI applications in the industry, which Cisco AI Defense aims to address with expertise from Robust Intelligence, a company acquired by Cisco’s Security Business Group. Robust Intelligence’s research paper on algorithmically assessing the security of large language models was integrated into Cisco AI Defense to enhance its security features.
The product will automatically propose security guardrails and continuously update them as models are refined and updated, preventing potential security risks that may arise from model adjustments. DJ Sampath, the vice president of AI software and platform in Cisco’s Security Business Group, emphasized the efficiency of the algorithmic approach to model validation compared to manual validation techniques. He noted that Cisco AI Defense can validate a model within 30 seconds, a process that typically takes companies seven to 10 weeks to complete manually.
Cisco AI Defense combines Robust Intelligence IP with large language models enriched by Cisco partner Scale AI, leveraging threat intelligence data from Talos and Splunk. Users can enforce guardrails through the Cisco cloud platform, with dashboards for AI Defense integrated into the Cisco Secure Access SaaS tool. Analysts have praised Cisco’s integrated approach to AI cybersecurity, especially its ability to provide ongoing guardrail updates and adapt to the evolving needs of enterprise IT as AI applications progress.
In the future, Cisco AI Defense plans to support on-premises workloads, edge computing, and IoT devices, as well as integration with Cisco’s Hypershield network security fabric. This integration will enable guardrail enforcement on east-west traffic, complementing the focus on north-south traffic where most large language model usage currently occurs. While Cisco’s expertise in networking infrastructure provides a strong foundation for AI cybersecurity, some analysts have raised concerns about the unproven nature of the Robust Intelligence IP and the complexities involved in deploying and running the new product.
Beth Pariseau, a senior news writer for Informa TechTarget, highlighted the significance of Cisco’s innovation in AI cybersecurity and the need for the company to demonstrate the effectiveness of Cisco AI Defense in real-world applications. Overall, Cisco AI Defense represents a step forward in enhancing AI cybersecurity for enterprise SecOps and protecting against potential threats within corporate networks.