Cisco Boosts AI Security for Enterprises

Cisco has introduced a comprehensive set of new security features designed to empower businesses to confidently deploy autonomous, or agentic, AI systems. This initiative addresses the critical need to protect these advanced AI agents, govern their complex interactions, and ensure resilient, secure connectivity as they operate across hybrid IT environments. The move reflects the shifting landscape where AI transitions from a passive assistant to an active agent capable of making decisions and taking actions using various tools and data sources.

The evolution from simple AI assistants to autonomous agents significantly broadens the potential attack surface. These agents interact with a wide array of tools, data stores, and external services, creating new vulnerabilities across the entire AI supply chain. Security teams now face the dual challenge of preventing the agents themselves from being compromised while also controlling what those agents are permitted to access and do on the organization’s behalf. As agents take on critical enterprise roles, we’re developing protections that work both ways: preventing agents from being compromised and controlling what they can access and do on our behalf, explained Jeetu Patel, Cisco’s President and Chief Product Officer.

To counter these emerging threats, Cisco is significantly expanding its AI Defense platform. The new capabilities focus on providing deeper visibility and stronger governance over the AI ecosystem. A key addition is the AI Bill of Materials (BOM), which offers centralized oversight of AI software assets, including model context protocol servers and third-party dependencies. This tool is crucial for securing the AI supply chain by helping teams understand the provenance of their AI components. Complementing this is an MCP Catalog for discovering and managing risks associated with MCP servers across public and private platforms.

Further strengthening security assessments, Cisco has introduced advanced algorithmic red teaming. This feature expands testing scope with adaptive, multi-turn evaluations for models and agents in multiple languages, simulating sophisticated attack scenarios. For real-time protection, the platform now includes agentic guardrails that continuously monitor interactions to detect manipulation, such as poisoned tools or malicious prompts designed to trigger unauthorized actions. These updates collectively help organizations inventory assets, understand their origins, and identify vulnerabilities much earlier in the development lifecycle.

The platform’s runtime protections have also been enhanced through an integration with NVIDIA NeMo Guardrails, providing a modular architecture for real-time security in production environments. This integration is part of the broader Cisco Secure AI Factory with NVIDIA, a validated reference architecture for powering AI workloads securely. Industry analysts note that these tools directly address the pressing questions security teams now face regarding asset inventory, provenance, and behavioral safety in production.

Beyond protecting the agents themselves, enterprises must ensure that the workflows powering these AI systems are both governed and reliable. AI agents depend on continuous, low-latency interactions with large language models, SaaS applications, and remote data endpoints. Performance issues or security gaps in these connections can halt critical processes. To meet this need, Cisco is enhancing its Secure Access Service Edge (SASE) offering with new AI-specific capabilities.

These include AI traffic optimization techniques, such as packet duplication, to maintain predictable performance during traffic surges. The platform also provides MCP visibility and policy control, allowing for the discovery and governance of agent-to-tool communications. Perhaps most critically, it introduces intent-aware inspection, which combines rapid detection with cloud-based analysis to evaluate the semantic intent behind agentic messages, going beyond what conventional security tools can analyze. Unified policy enforcement across software-defined wide area networking and security service edge components simplifies governance as AI adoption accelerates.

Finally, as agentic AI becomes embedded in mission-critical operations, the underlying network infrastructure must support secure, encrypted connectivity at scale. Cisco addresses this with the announcement of IOS XE 26, the latest version of its core networking operating system. This release powers its 8000 Series Secure Routers and C9000 Series Smart Switches, and introduces new router variants for small and mid-size businesses.

A cornerstone of IOS XE 26 is its implementation of full-stack post-quantum cryptography (PQC). This advanced encryption is designed to protect organizations against future threats from quantum computing, helping to defend against device tampering and data compromise. This proactive step aligns with evolving global regulatory guidance and ensures that encrypted AI communications remain confidential for the long term. Together, these networking advancements help maintain predictable performance for AI traffic across distributed locations while future-proofing security from the network core out to campus and branch environments where AI workflows are increasingly initiated.

(Source: HelpNet Security)