AI agents have access to key data across the enterprise
82% of organizations already use AI agents, but only 44% of organizations report having policies in place to secure them, according to SailPoint. While 53% are in the process of developing such policies, the reality is that most remain exposed today.
AI agents pose security risks for organizations
96% of technology professionals consider AI agents a growing risk, even as 98% of organizations plan to expand their use of them within the next year.
The terms “AI agent” or “agentic AI” broadly encompass autonomous systems that perceive, make decisions, and take action to achieve specific goals within an environment. These agents often require several different machine identities to access needed data, applications and services, and they introduce additional complexities like self-modification and the potential to generate sub-agents.
72% of technology professionals now believe AI agents present a greater risk to the business than traditional machine identities.
Factors contributing to AI agents as a security risk include:
- AI agents’ ability to access privileged data (60%)
- Their potential to perform unintended actions (58%)
- Sharing privileged data (57%)
- Making decisions based on inaccurate or unverified data (55%)
- Accessing and sharing inappropriate information (54%)
“Agentic AI is both a powerful force for innovation and a potential risk,” said Chandra Gnanasambandam, EVP of Product and CTO at SailPoint. “These autonomous agents are transforming how work gets done, but they also introduce a new attack surface. They often operate with broad access to sensitive systems and data, yet have limited oversight. That combination of high privilege and low visibility creates a prime target for attackers. As organizations expand their use of AI agents, they must take an identity-first approach to ensure these agents are governed as strictly as human users, with real-time permissions, least privilege and full visibility into their actions.”
Governing AI agents
Today, AI agents have access to customer information, financial data, intellectual property, legal documents, supply chain transactions, and other highly sensitive data. Yet respondents reported deep concerns over the ability to control the data AI agents can access and share, with an 92% stating that governing AI agents is critical to enterprise security.
23% reported their AI agents have been tricked into revealing access credentials. Additionally, 80% of companies say their AI agents have taken unintended actions.
Leading the list of unintended actions, 39% of respondents reported AI agents accessed unauthorized systems, while 33% said agents accessed inappropriate or sensitive data. Although these behaviors may reflect attempts to fulfill a task, the next set of actions is more concerning: 32% noted that AI agents enabled the download of sensitive data, and 31% said the data was inappropriately shared.
IT stands out in AI agent data access knowledge
IT (71%) is the most informed team regarding AI agent data access, given their role in implementing the technology, managing configurations, and provisioning credentials.
However, awareness drops significantly among other critical stakeholders: compliance (47%), legal (39%), executives (34%), and other departments—despite their role in identifying sensitive data, safeguarding the organization, and minimizing risk.
This lack of visibility into data access has resulted in only 52% of companies reporting that they can track and audit all data used or shared by AI agents.
Despite needing multiple permissions to perform their functions, AI agents typically receive expedited access through IT departments alone. By design, these agents will explore all accessible resources to fulfill requests—creating inherent security vulnerabilities.
To address these risks, organizations must implement specialized identity security solutions with AI agent-specific controls that can restrict access to sensitive data, maintain comprehensive audit trails, and provide transparency to all stakeholders.