AWS re:Invent 2025 - AI Agents – the new face of privileged machine identities (SEC226)

Summary

In this session, Venu Shastri and Anat Eytan-Davidi from CyberArk explore the growing importance of securing AI agents in the enterprise. As AI agents become more integrated into business processes, their elevated access privileges create new security risks. The discussion covers the unique challenges of managing AI agent identities, including the need for least privilege access, real-time monitoring, and dynamic policy enforcement. CyberArk introduces its Secure AI Agent Solution to address these risks, offering tools like access control, audit tracking, and secure credentials. The session stresses the urgency of securing AI agents to minimize potential threats in the future.

Takeaways

• 😀 AI agents are rapidly becoming the new face of privileged machine identities in enterprises, offering the potential to unlock significant business value.
• 😀 Securing AI agents is crucial as they are given elevated privileges to access enterprise resources, which introduces unprecedented risks, including unintended actions by the agents themselves.
• 😀 Early-stage AI agents are prone to hallucinations and misinterpretations, which can lead to unintended actions that could compromise security.
• 😀 As AI agent adoption grows, risks multiply due to the increasing complexity of their roles and access to enterprise resources, requiring a focus on securing them at every stage of development.
• 😀 AI agents need to be treated as privileged machine identities, combining the best practices of both human identity and machine identity security to ensure comprehensive protection.
• 😀 Traditional MFA and static access controls are inadequate for securing dynamic, complex AI agents. Context-aware dynamic policies are necessary to handle the complexity of their access.
• 😀 The concept of least privilege must be enforced for AI agents, ensuring they only have access to necessary resources at the right time, reducing the potential attack surface.
• 😀 AI agents require a robust auditing and monitoring framework to track their actions and ensure that any unintended or unauthorized behavior can be quickly detected and addressed.
• 😀 CyberArk's Secure AI agent solution combines human and machine security best practices, focusing on zero standing privileges, session monitoring, and strong authentication methods.
• 😀 The industry is witnessing a shift towards autonomous AI agents that operate at machine scale and speed, making the need for secure AI agent management even more critical.
• 😀 The integration of tools like CyberArk's AI Agent Gateway and its dynamic, context-aware policies helps secure AI agents' access, control permissions, and ensure compliance across different user personas and access needs.

Q & A

What are AI agents, and why are they becoming increasingly important in enterprises?

AI agents are autonomous systems powered by large language models (LLMs) that perform tasks and make decisions on behalf of users or organizations. They are becoming crucial in enterprises because they can automate complex processes, unlock significant business value, and manage tasks at machine scale and speed, which is critical for modern businesses.

What are the risks associated with giving AI agents elevated privileges to access enterprise resources?

The main risks include AI agents potentially making unintended actions due to their 'hallucinations' or misunderstanding of context, as well as the possibility of these agents being manipulated by external threat actors. As their adoption grows, the complexity and risks increase, especially as AI agents access more sensitive resources.

How does the growth in AI agent adoption impact enterprise security?

As AI agents are given more privileges and access to enterprise resources, the attack surface grows. The rapid development of AI agents leads to an increase in complexity, which makes it harder to control and secure them. This makes it essential to establish robust identity security practices early on to prevent potential vulnerabilities.

What is the new identity security standard for AI agents, and how does it relate to traditional access control methods?

AI agents are now considered privileged machine identities, which require different access controls compared to traditional human identities. While traditional MFA and access policies may not be sufficient, AI agents require dynamic, context-aware policies to ensure they operate securely. These policies must factor in the user, the AI agent itself, the risk, and the context of the request.

What are the core components of CyberArk's Secure AI Agent solution?

CyberArk’s Secure AI Agent solution focuses on four key areas: discovery and context, securing access through least privilege and zero standing privileges, threat detection and response, and governance to ensure compliance. The solution aims to safeguard AI agent operations across different environments, ensuring they don’t perform unintended or risky actions.

Why is it critical to use identity security as the foundation for securing AI agents?

Identity security provides the foundation for ensuring AI agents are granted only the necessary permissions and operate with minimal risk. By implementing least privilege, zero standing privileges, just-in-time access, and strict monitoring, enterprises can secure AI agents and reduce potential attack surfaces.

What role do 'least privilege' and 'just-in-time' access play in securing AI agents?

‘Least privilege’ ensures AI agents have only the minimum permissions necessary to perform their tasks, while ‘just-in-time’ access grants those permissions only when needed, and revokes them afterward. These practices help prevent excessive permissions, which could lead to a security breach if compromised.

How does CyberArk’s AI Agent Gateway help secure access to AI agents?

The AI Agent Gateway acts as an enforcement point between the AI agent and the resources it can access. It ensures that AI agents follow least privilege principles, enforces human-in-the-loop control when needed, and provides traceability for all actions performed by the AI agent, ensuring that any unauthorized or unintended actions are detected.

What are the key security risks associated with storing API keys and secrets in environment variables, and how does CyberArk address these risks?

Storing API keys and secrets in environment variables exposes sensitive information to potential leakage, especially if the environment is compromised. CyberArk’s open-source security toolkit addresses this by ensuring credentials are injected just-in-time, and removed when no longer needed, reducing the risk of memory exposure and unintended leakage.

What can organizations do today to start securing their AI agents?

Organizations should begin by thinking about identity security for AI agents, ensuring they implement least privilege access, zero standing privileges, and just-in-time access. Additionally, businesses should monitor AI agent behaviors and begin using solutions like CyberArk’s Secure AI Agent solution to gain better visibility, control, and governance over AI agent access and actions.