The numbers arriving from enterprise environments are striking enough to shift the conversation about AI risk. Non-human identities in Fortune 500 companies grew 480% in the past six months, according to data cited by Cyera, which launched Agent Guardian at Black Hat 2026 this week to address the gap between how quickly AI agents are being deployed and how little visibility organisations currently have over what those agents are doing.
AI agents present a different kind of access problem to traditional software. When an agent is deployed, it typically inherits the elevated permissions of the human user or service account that launched it, then operates at machine speed — reaching data in seconds that a human might take days to access. Shadow agents, those deployed without formal security review, complicate the picture further. Security teams frequently have no record they exist.
Agent Guardian organises the response around four stages: discovery, governance, protection, and validation. Discovery inventories every AI application and agent running across endpoints, SaaS platforms, and cloud environments, including shadow agents and MCP servers. Governance defines what each agent is permitted to do and flags when it drifts from its intended purpose. The protection layer operates inline — intercepting dangerous tool calls, blocking unauthorised data transfers, and quarantining compromised agents before execution. Validation red-teams defences against prompt injection, jailbreaking, and privilege escalation, and generates audit-ready compliance reports for frameworks including the EU AI Act.
The platform covers the breadth of the current agentic enterprise: enterprise deployments on AWS Bedrock, Microsoft Copilot Studio, Salesforce Agentforce, and Snowflake Cortex AI are all in scope, as are local developer agents such as Claude Code and Cursor, which operate on employee devices outside corporate network controls. For that last category, Cyera introduced a companion product: Cyera Endpoint, a lightweight agent that brings the same governance policies to the device level.
"AI agents have become a new class of enterprise identity. But the fundamental shift isn't that organisations have more identities; it's that the gap between permission and execution has disappeared," said Tamar Bar-Ilan, cofounder and CTO of Cyera. "Agent Guardian and Cyera Endpoint give organisations the visibility and runtime control to govern this new reality, enabling them to accelerate AI adoption without putting sensitive data at risk."
Snowflake is among Cyera's early integration partners. "Working with Cyera lets Snowflake customers see every Cortex AI agent in use and understand exactly what data it can reach, so they can adopt agentic AI with the same confidence they place in their data," said Mayank Upadhyay, Chief Security and Trust Officer at Snowflake.
To stay across the latest in cloud, AI and enterprise tech analysis from Compare the Cloud, subscribe to our weekly newsletter at https://www.comparethecloud.net/newsletter
