As organizations rapidly deploy AI agents to automate tasks and increase efficiency, a new powerful and privileged identity class emerges. These autonomous entities bring new risks, such as inaccurate or incorrect agent output (hallucination), misuse, and the potential for malicious actors to take them over. These risks are amplified when agents have elevated access rights.
The CyberArk Secure AI Agents solution addresses these challenges by applying the right level of access management. This ensures that AI agents only have access to what they need, precisely when they need it, and no more. This approach reduces risk, helps prevent unauthorized access, and enables organizations to scale AI-driven initiatives safely and confidently.
"As organizations deploy AI agents, both developers and security teams must be aware of the changing identity risks," said Matt Cohen, CEO of CyberArk. "Without robust detection, onboarding, access management, and lifecycle management, organizations lose visibility and control, creating room for shadow AI (AI use outside of IT's view) and unauthorized access. Only CyberArk secures the full spectrum of identities, from people and machines to AI agents, allowing organizations to innovate without compromising security and compliance."
Security of AI Agents requires a new approach
According to new research from CyberArk, the use of AI agents is expected to rise to 76% over the next three years, while less than 10% of organizations have adequate security measures in place. The research 'Securing Agentic AI: Identity as the Emerging Foundation for Defense' shows that fewer than one in ten organizations have implemented security measures for agents at scale. Two-thirds of CISOs in the financial services and software sectors view agentic AI as one of the three biggest cybersecurity risks, with a third naming it their top concern. Yet nearly 40% of large financial institutions and software companies have already deployed agentic AI in production. Most expect that securing AI agents will lead to increased spending in the coming year.
Privilege controls for every identity
AI agents represent a new category of privileged machine or non-human identities that mimic human reasoning and decision-making. While they accelerate productivity, they also introduce new risks. These include hallucinations, autonomous misuse, and takeovers of AI agents by cybercriminals. AI agents often have access to sensitive systems and data, necessitating a security solution that combines capabilities for both human and machine identities.
The key features of the CyberArk Secure AI Agents solution are:
- Comprehensive agent detection: Automatic detection of all AI agents in SaaS, cloud, and development environments, with extensive profiles including owners, roles, and access rights.
- Secure agent access: Enforcement of strong authentication and least-privilege access, with zero standing privileges and complete control over agent activities.
- Real-time threat detection: Continuous monitoring of anomalies and unauthorized access, with automatic alerts and rapid response.
- Lifecycle management and compliance: Management of each AI agent from creation to decommissioning, with support for regulations and audit preparation.
CyberArk Secure AI Agents will be available in December 2025.
