Ninety-one percent of organizations are now running AI agents inside their enterprise systems. But there is a startling blind spot: only 10 percent have a developed strategy for managing the identities of those agents. That gap is the next major security frontier in AI adoption.

AI agent identity management in 2026 is no longer a niche security concern. It is quickly becoming the defining infrastructure challenge for enterprises deploying agentic AI at scale. When an AI agent logs into your CRM, queries your database, or sends a message in your name, does your IT team know which agent did it, why, and with what permissions? In most organizations, the honest answer is no.

This post breaks down why AI agent identities matter, what is driving a new wave of dedicated solutions from companies like NewCore and AppViewX, and what practical steps your organization should take right now to govern your growing AI workforce.

The Non-Human Identity Security Crisis No One Saw Coming

The security industry spent decades solving human identity. Multi-factor authentication, single sign-on, role-based access control: these tools were built for people. AI agents operate on entirely different terms.

Unlike a human employee who logs in once per day, an AI agent may authenticate hundreds of times per hour across dozens of systems. It runs continuously, often across cloud environments, internal applications, and third-party APIs simultaneously. It acquires permissions incrementally, sometimes in ways its original developers never anticipated. And it generates activity at machine speed, making manual auditing nearly impossible.

Okta’s “AI Agents at Work 2026” report captures the scale of the problem: 50 percent of identity activity in enterprise systems is now invisible to central IAM teams. The firm calls this “identity dark matter.” Two out of three non-human accounts, including AI agents, are set up locally within individual applications, completely outside central governance.

This non-human identity security gap is not a theoretical risk. It is already showing up in compliance audits, access control failures, and agentic workflows that quietly accumulate permissions over time. Nearly two-thirds of organizations apply weaker security controls to AI agents than to human employees, according to IANS Research.

The core insight from every major security vendor working on this problem is the same: AI agents must be treated as first-class identities, with the same lifecycle management, granular permissions, and revocation controls as any employee. For a deeper look at the broader threat landscape, see the full breakdown of AI agent security risks every business needs to know.

NewCore, AppViewX, and the Race to Solve Enterprise AI Agent Security

A new market is forming around enterprise AI agent security specifically, and it is moving fast.

NewCore, founded by Zohar Alon (previously of Check Point-acquired Dome9), emerged from stealth in June 2026 with $66 million in funding from Index Ventures, Cyberstarts, and Evolution Equity Partners. NewCore’s platform treats every AI agent like an employee: it provides managed identities, granular permissions, complete audit logs, and instant revocation controls. The system integrates directly with coding assistants like Anthropic’s Claude Code, OpenAI’s Codex, and Cursor, allowing those tools to access enterprise systems as managed identities rather than through manually distributed credentials. TechCrunch covered the launch in detail here.

AppViewX launched its own “Agent Identity Security” product around the same time, taking a PKI-based approach to discovering, governing, securing, and monitoring AI agents across enterprise systems. Rather than requiring a complete IAM overhaul, AppViewX targets the specific gap between existing identity infrastructure and the new non-human identity sprawl created by agentic deployments.

At the platform level, Cisco announced its agentic workforce identity model at RSA Conference 2026, offering a six-stage maturity framework for AI agent IAM governance. Microsoft extended its Agent 365 control plane specifically to give IT teams visibility over agents running across Microsoft 365 and Azure environments. Okta’s comprehensive AI Agents at Work 2026 report ties all these trends together with enterprise survey data.

The common thread: every serious enterprise AI vendor now recognizes that ungoverned AI agents are a security liability, and purpose-built identity tools are the answer. This mirrors broader trends in AI agent governance for enterprise teams, where proportional controls are proving far more effective than blanket policies.

How to Manage AI Agent Identities in Enterprise Systems

For organizations building or expanding agentic AI deployments, the practical path to managing AI agent identities starts with three foundational steps.

Inventory first. You cannot govern what you cannot see. Before adding governance controls, run a discovery pass across your cloud environments, SaaS applications, and internal systems to identify every non-human identity currently operating. Treat any API key, service account, or OAuth token connected to an autonomous workflow as a candidate for formal agent identity enrollment. Forrester estimates that 30 percent of enterprise app vendors will publish MCP servers in 2026, each of which creates new agent connectivity points that need tracking.

Apply a tiered permission model. Not every agent needs admin credentials. Use the principle of least privilege: assign each agent the minimum permissions required for its specific task, set expiration dates on elevated access, and use tools like NewCore or AppViewX to enforce these policies dynamically. Organizations that have adopted proportional governance, matching security controls to the autonomy level of each agent, report significantly lower incident rates than those using uniform controls.

Audit continuously, not periodically. AI agents act at machine speed. Quarterly audits of human accounts are already inadequate for detecting misuse; for AI agents, they are essentially useless. Implement real-time activity monitoring tied to each agent’s formal identity so any anomalous behavior triggers an immediate alert.

These three steps apply whether you have five agents deployed or five thousand. The earlier you build the governance scaffold, the less painful the scaling becomes. Teams just starting their agentic journey may find the AI agents for small business guide useful for context on lower-stakes starting points before expanding to larger deployments.

The Future of AI Agent Identity Goes National

The significance of AI agent identity is not lost on policymakers. Estonia, globally recognized as the leader in digital governance, announced in 2026 that it will issue official digital identity codes to AI agents operating within its national systems. It marks the first time a nation-state has treated AI agents as distinct legal identity entities rather than extensions of the humans or companies that deploy them.

At the enterprise level, the trajectory is equally clear. Gartner projects that by 2027, formal AI agent identity frameworks will be a procurement requirement for 40 percent of enterprise software contracts. Major compliance standards, including SOC 2, ISO 27001, and PCI DSS, are actively working to operationalize non-human identity requirements, following the Cloud Security Alliance’s publication of the NIST AI RMF Agentic Profile.

The pattern repeats what happened with every prior infrastructure shift: first adoption, then security incident, then governance standard. With AI agents, the industry is trying to compress that cycle before the incident phase becomes widespread. Organizations that invest in AI agent identity management now will have a significant compliance and security advantage as standards tighten through 2027.

Conclusion

Three things are clear about AI agent identity management in 2026. First, the governance gap is real and growing: 91 percent adoption with only 10 percent formal strategy is not a sustainable ratio. Second, dedicated solutions now exist, with NewCore, AppViewX, Cisco, and Microsoft all launching purpose-built capabilities this year. Third, regulatory standards are coming: Estonia’s national AI ID codes and Gartner’s 2027 procurement prediction signal that governance will shift from best practice to baseline requirement.

The organizations that treat AI agent identity as a first-class infrastructure problem today will be the ones scaling agentic AI with confidence tomorrow.

For more guides on deploying AI agents responsibly and effectively, explore the full resource library at BigAIAgent.tech. What is your organization’s current approach to governing AI agent permissions? Share your experience in the comments.