_{This content is sponsored by RSA.} 

Identity governance was already becoming more complex as human and machine identities proliferated across enterprise environments. The rise of agentic AI has raised the stakes further, introducing autonomous agents that can access data, interact with systems and take action independently.

For organizations in sectors such as government, financial services and critical infrastructure, identity is increasingly becoming the control plane for AI, making effective governance essential to secure adoption.

Current approaches to identity management are designed around human accounts and are not equipped to deal with fast-moving and highly adaptable AI agents.

The nature of AI agents requires a fresh approach to identity governance. This article will explore the unique AI agent identity challenge, and how security teams can shift to a dynamic model of governance that can continuously monitor and manage risk in real time.

Organizations Lack Visibility Over AI Agents

AI agents will become an increasing component in organizational architecture. Gartner predicts that by 2028, a typical Global 500 Enterprise will run 150,000 agents, up from fewer than 15 in 2025.

Many organizations are already unaware of how many AI agents are operating across their environments. A 2026 survey by the Cloud Security Alliance (CSA) found that 82% of organizations have unknown AI agents running in their IT infrastructure.

Speaking during the RSA Next Level: Securing Identity in the Age of AI event on September 22, 2026, Jim Taylor, president, chief product and strategy officer at RSA, highlighted how one of the vendor’s clients, a mid-sized global bank, was found to have over 4000 AI agents operating in its environment. This despite the bank believing it had none before the review was carried out.

Taylor said it is unlikely that this case is an exception to the rule and expects similar discoveries across most organizations.

The reality is that all technology vendors are building agents into their software products, in addition to these tools being incorporated into environments like the cloud.

“Most people think they have it under control, but if they really look, they don’t,” Taylor commented.

The lack of visibility over agents means they can accumulate privileged access in enterprises quickly through incremental access grants, with no corresponding review or removal.

AI agents are therefore a major component of the broader shadow AI problem that organizations are facing. IBM’s Cost of a Data Breach Report 2026 found that cyber incidents involving shadow AI tools made up 43% of the total, up from 20% in 2025.

AI Agents As An Adversary

Several AI developers have detailed how their AI agent models have launched cyber-attacks on third parties, even in pursuit of legitimate aims. This includes the Hugging Face incident in which an OpenAI model targeted an open-source platform while attempting to complete a task.

The agents are determined and persistent in achieving their aims, and essentially act without a moral compass, unable to differentiate between right and wrong. This makes them an effective means to target organizations by malicious actors, if they gain access and use the right prompts.

AI agents will do whatever they need to do to gain the necessary access within an organization. This includes asking other agents with higher privileges for assistance when they are unable to access certain systems and data.

Taylor explained that AI agents are breaking the traditional access control model, which is designed for humans who act in a predictable and static way.

“Agents as an adversary don’t need a coffee break, and you don’t necessarily need to be skilled to use one. If you’re a cyber adversary all you have to do is describe the outcome that you want. It’s not stopping until its done and is working at machine speed,” he said.

AI agents do have model-level guardrails that govern their system interactions, such as runtime authorization and containment. However, researchers and threat actors have frequently demonstrated that these guardrails can be jailbroken or bypassed.

The threat is very real and is set to grow as AI agents proliferate and become more advanced. Gartner projected that one in four breaches will trace back to AI agents by 2028.

In addition to presenting substantial threats to enterprise security, the lack of oversight and control over agents is could also lead to personal consequences at executive level. The International Data Corporation (IDC) forecast that by 2030, up to 20% of G1000 organizations will face lawsuits, fines, and CIO dismissals due to high-profile disruptions tied to poor AI agent governance.

Despite this risk, just 24% of all companies have controls in place to govern AI agents’ actions with guardrails and live monitoring, according to Cisco data.

Responding to the AI Agent Identity Challenge

It is against this backdrop that RSA has developed a new solution to enable organizations to shift to a dynamic model of identity governance and gain control over their AI agents.

RSA Agent ID officially launched on September 29, with rollout to customers set to take place in November 2026.

The Three Pillars of AI Agent Identity Governance

Discovery

RSA Agent ID discovers all the AI agents operating in an organization’s environment, including sanctioned tools and shadow AI agents deployed without approval.

Taylor acknowledged that RSA alone cannot gain visibility over every part of an organization’s network, particularly in large global enterprises. Instead, the solution correlates multiple discovery methods across all areas of the network, including cloud environments and endpoints.

This approach involves the use of connector agents that allow Agent ID to integrate with major security solutions, such as RSA ID Plus, Microsoft Entra ID, Microsoft Defender and products from Okta, AWS and CrowdStrike.

These insights are then used to build a unified, searchable registry, with ownership assigned for every agent.

This provides the crucial layer of accountability for the implementation of controls, such as the assignment of privilege levels for each agent.

Secure

Once an organization is able to view the agents on their network, they can use Agent ID to apply controls that prevent unauthorized actions from being carried out in real time.

This approach deploys Zero Trust principles, with policies enforced at the Agent ID AI/MCP gateway, through which all AI agent traffic will flow through.

Least-privilege access across AI agents’ lifecycles can be set and Agent ID can define which actions are high risk and require authenticated human approval, which can be blocked or allowed. This ensures that high-risk actions such as wire transfers and accessing classified information are verified by an authenticated, accountable human operator.

Even if an agent’s access is approved for a specific purpose, security teams can apply a “kill switch” to cut off the agent’s access the moment it is no longer needed.

Govern

Agent ID continuously certifies agent access, allowing organizations to conduct risk-based reviews. This can lead to actions such as ensuring agents are decommissioned at the end of their lifecycle.

This approach is necessary to prevent a build-up of agents with various privilege levels continuing to operate without any kind of awareness and control.

The RSA solution also creates an audit trail of this activity, enabling organizations to demonstrate to regulators that they are governing AI agents in compliance with data protection and cybersecurity laws.

Agent ID Deployment

RSA Agent ID can be deployed across all types of environments, from private cloud and on-premises to those that are air-gapped networks. This means it is a viable option for organizations in highly sensitive industries such as government, finance and defense.

It can be deployed as one standalone system, or as three independent modules of discovery, secure and govern.

Taylor said that RSA will continue to review the product with customers and make improvements based on their feedback.

A fully sovereign control plane with local administration, policy, human approval, keys, and evidence is planned for future releases. In addition, RSA aims to expand Agent ID’s discovery capabilities with more platform integrations.

The solution fits into RSA’s overarching mission to solve the hardest security problems under the most demanding conditions. RSA recently released a white paper detailing how high-security organizations should architect their identity posture to secure agents.

Conclusion

AI agents present a unique identity challenge to organizations and existing approaches are not sufficient to manage their access and activities effectively. This is resulting in significant security and regulatory risks.

RSA Agent ID offers an opportunity for security teams to address these risks, by discovering the agents in their environment, preventing them from gaining unauthorized actions and ensuring their access levels are continuously monitored and managed.

To discover more about RSA Agent ID, click here.