Shine a light on shadow AI—agentic security starts with visibility
Organizations across the spectrum are under immense pressure to move fast and at incredible scale—often at the expense of safety and security. However, in order to win in the agentic age, organizations are going to have to find a way to unleash AI agents with the appropriate guardrails. This, of course, starts with agentic visibility.
Cisco today announced the launch of AI Discovery in Cisco Identity Intelligence, a solution that gives organizations a complete record of everything about an agent and associated resources. This includes the full picture of where agents are running, what they are connected to, what credentials they use, and what they can access.
The need to understand agent exposure
Despite amazing progress at incredible velocity, we're still in the wild west of agentic adoption. Developers, sales teams, marketers—they're all moving forward at different speeds, experimenting with new models and tools to create operational efficiencies that help them do more work, better, and in less time. All these tools are decentralized across the organization, making it hard to keep track of who is doing what—much less how these agents should be governed.
The old adage that you can't secure what you can't see certainly applies to AI agents. But it's not just about what you don't know or can't monitor. The seemingly endless ways in which agents move between business systems, connect to outside resources, and access proprietary information create massive identity sprawl that expands threat surfaces and increases the blast radius of an incident. According to SpecterOps, 40,000 identities can lead to 725 million attack paths, giving attackers infinite ways to gain access and spread laterally across networks and business systems.
The risks are more than hypothetical. An agent's access can exceed its task. A shared assistant may use its author's permissions. A local coding agent may connect to an overlooked Model Context Protocol (MCP) server. An inactive agent may retain access it no longer needs. All these present dangerous security risks for the organization, and evidence behind these actions sits across dozens, if not hundreds, of systems—including Software as a Service (SaaS) configurations, endpoint telemetry, and identity and access management solutions.
Turning identity sprawl into identity context
AI Discovery in Cisco Identity Intelligence gives organizations a complete record of everything about an agent and associated resources. This includes the full picture of where agents are running, what they are connected to, what credentials they use, and what they can access. Essentially, it does what you are likely already doing today for human identities, but with purpose-built identity context for machines and agents. This agentic context can then be used to augment critical security functions—such as threat detection, network segmentation, and regulatory compliance—as agents and humans continue to collaborate in new, innovative ways across the enterprise.
Gaining a full picture of agentic operations through Cisco allows security teams to better understand their exposure from agent activity across the organization. Continuous authentication checks ensure all configurations of these dependencies are reviewed in near real time, identifying potential abnormalities that teams can prioritize. At the same time, the agent inventory gives security and compliance teams a shared basis for review. Organizations can easily explain to auditors what was found, identify security gaps, and combine the findings with their own governance records.
Creating a central agent inventory
Cisco's AI agent discovery architecture is built around where agents actually leave traces, not just where they're registered. We do this in two distinct methods:
Platform integrations: Cisco connects directly to AI vendor APIs—such as Microsoft Copilot, Amazon Bedrock, Google Vertex, OpenAI, Salesforce Agentforce, and other major agentic platforms. Every registered agent and MCP server across these environments is captured in real time.
Sensor telemetry: Cisco reads signals from EDR solutions, network sensors, and browser extensions to surface agents and MCP servers running locally on managed devices—including agents that never communicate with a cloud platform.
Get a free agent exposure assessment
Agentic adoption isn't slowing down. Nor is it going to inherently follow established security policies and controls. Cisco brings agentic visibility to bear, creating a central inventory of AI agents that shed light on your organization's AI usage. This information is put into business context, presented visually for reporting and compliance, and fed into other tools to help bring identity context to detection, response, segmentation, and other critical security functions.
Importantly, organizations won't have to overhaul their identity or security infrastructure to gain agentic visibility. The solution is available in Duo Advantage at no additional cost and comes with pre-existing integrations built right in. We've even built a free tool that security teams can use to get a snapshot of agent exposure in their environments.
Take the agentic visibility assessment now and immediately get a better handle on Agentic AI adoption in your organization.