Key Takeaways:
- 71% of CISOs say AI has access to core business systems, but only 16% govern that access effectively.
- 92% of organizations lack full visibility into AI identities, and 95% doubt they could detect misuse if it happened.
- 86% don’t enforce access policies for AI identities, and only 17% govern even half of their AI identities like human users.
- 75% have discovered unsanctioned AI tools currently running in their environments, often with embedded credentials or elevated system access that no one is monitoring.
- Only 25% use AI-specific identity or monitoring controls, leaving most organizations vulnerable to AI-related security risks.
Introduction to the AI Security Challenge
The rapid expansion of Artificial Intelligence (AI) in business environments has created a significant security challenge for organizations. Many security leaders have found that AI systems have been introduced into their environments without their explicit approval, often with elevated access levels and autonomy. This has resulted in a lack of visibility and control over AI activities, making it difficult for security teams to answer basic questions such as "Who did this?" and "Should this action have been allowed?" A recent survey of over 200 CISOs and security leaders revealed that 71% of organizations have AI tools with access to core business systems, but only 16% govern that access effectively.
The Challenge of AI Identities
AI identities don’t behave like human users or traditional service accounts, making it difficult for security teams to manage them using traditional Identity and Access Management (IAM) tools. These AI identities can act independently, create additional identities, and escalate privileges without permission, making them a significant security risk. The survey found that 92% of organizations lack full visibility into AI identities, and 95% doubt they could detect misuse if it happened. Furthermore, 86% don’t enforce access policies for AI identities, and only 17% govern even half of their AI identities like human users. To address this challenge, security teams need to think differently about AI identities and develop a paradigm shift in how they govern them.
The Visibility Crisis
Before security teams can control AI access, they need to know where these agents are operating. However, most organizations lack full visibility into their AI identities, making it difficult to detect and respond to security incidents. The survey found that 92% of organizations lack full visibility into AI identities, and 95% doubt they could detect misuse if it happened. To address this crisis, security teams need to invest in AI-specific identity and monitoring controls, such as API and workload identity discovery and inventory, posture analytics, and continuous monitoring. By doing so, they can gain the visibility and control needed to manage AI-related security risks.
The Weakest Link: AI Governance
AI tools are now acting with real authority and privilege, but most organizations lack effective governance policies to manage them. The survey found that 86% of security leaders lack or don’t enforce access policies for AI identities, and only 19% govern even half of their GenAI accounts with the same rigor they apply to human users. To address this weakness, security teams need to develop governance policies that account for the unique characteristics of AI identities, such as their autonomy and ability to escalate privileges. They also need to apply a unified system of lifecycle, least-privilege, and certification controls to both human and AI identities.
The Shadow AI Problem
In many organizations, AI isn’t just being rolled out by IT through sanctioned projects. Instead, "Shadow AI" is showing up through unapproved AI tools that teams bring in on their own. The survey found that 75% of CISOs have discovered unsanctioned GenAI tools already running in their environments, often with embedded credentials or elevated system access that no one is monitoring. To address this problem, security teams need to help stakeholders implement AI tools in a safe and controlled manner, by setting up a small governance group, defining a straightforward review process, and evaluating each tool for access scope, embedded credentials, and data exposure.
The Limitations of Legacy Tools
Most identity tools were designed for human users, not autonomous AI systems. Many organizations are still trying to manage AI risk with tools designed for a different era and for environments comprising on-premises systems, human users, and static access. The survey found that only 25% of organizations use AI-specific identity or monitoring controls, leaving most organizations vulnerable to AI-related security risks. To address this limitation, security teams need to invest in AI-specific identity and monitoring controls, such as unified identity platforms that converge IGA, PAM, and access analytics.
The Path Forward: Identity as the Enforcement Layer
Perimeter controls don’t follow AI into cloud platforms, and device policies don’t apply to headless agents. Identity is the most consistent enforcement layer that remains where access decisions, privilege boundaries, and audit trails converge across environments. Security teams are shifting their priorities accordingly, investing in identity discovery and inventory, continuous monitoring and posture analytics, and AI-driven detection. By doing so, they can lay the groundwork for an identity-first approach, where identity becomes the consistent enforcement point for AI, verifying who or what is acting, with what privilege, and for how long.
Conclusion
The AI era requires a different approach to security, one that takes into account the unique characteristics of AI identities and their potential security risks. Organizations need to develop a new system that can evolve and scale just as quickly as AI, with a focus on identity security, governance, and monitoring. By doing so, they can enable their organization to leverage AI securely and use it as a strategic growth lever. The survey found that CISOs are moving toward always-on identity governance of AI operations, detecting privilege drift, enforcing policy in real time, and remediating risk automatically. This shift is happening now across the enterprise, and organizations that move first stand to gain a competitive advantage.