The expansion of agentic AI infrastructure inside modern enterprise networks is accelerating at an unprecedented rate, signaling a fundamental transformation in how business systems operate. Speaking at the FutureSec conference in Dubai, Mortada Ayad, Vice President of Sales for META at Delinea, warned that organizations could deploy as many as 46 AI agents for every human employee within six months. As businesses rapidly transition from basic conversational models to autonomous decision-making software, managing digital identities and controlling system privileges are becoming as critical as securing human workforce access.
Bottom-Up Adoption and the Five Stages of Autonomy
Unlike traditional digital transformation initiatives that follow top-down executive roadmaps, the current expansion of agentic AI infrastructure is driven from the bottom up. Employees and department heads are actively integrating autonomous AI tools directly into daily workflows to streamline operations and enhance productivity. Experts outline a clear five-stage progression for enterprise AI implementation: initial conversational chatbots, task-oriented copilots, autonomous agents, multi-agent collaborative networks, and fully autonomous operational systems. Modern enterprises are rapidly entering the multi-agent phase, embedding non-human decision-makers directly into internal data pipelines.
This decentralized adoption model creates significant management hurdles for IT and security leadership. Because individual teams deploy software agents independently, many organizations struggle to maintain an accurate inventory of active AI agents, their underlying access permissions, or their interaction boundaries with core corporate databases.
Re-evaluating Perimeter Defense and Trusted Internal System Threats
For decades, enterprise security strategy centered on protecting network perimeters, hardening endpoints, and monitoring external attack vectors. However, the expansion of agentic AI infrastructure forces security leaders to re-examine potential internal threats. When an AI agent possesses broad access to financial records, customer data, and proprietary databases, an improperly configured or compromised agent can cause as much operational disruption as a malicious external threat actor.
To mitigate these emerging security vulnerabilities, cybersecurity experts advocate for the VET AI framework, which treats AI agents with the same rigorous privilege protocols applied to high-level human executives:
Validating the Mission: Defining precise operational boundaries, objectives, and limitations for every active agent before deployment.
Establishing Identity and Authority: Assigning unique digital credentials, explicit permission scopes, and clear legal accountability to non-human entities.
Testing Systems: Conducting continuous stress testing, scenario simulations, and code audits to identify unpredictable behavior before live execution.
Ensuring Monitoring and Containment: Implementing real-time behavioral tracking, automated kill switches, and strict containment barriers to halt unauthorized system actions.
Identity Management as the New Security Perimeter
As digital agents gain authority to execute complex transactions, update codebases, and manage supply chain communications, identity and access management (IAM) is becoming the foundation of enterprise security. Treating autonomous software entities as privileged non-human identities ensures that access rights strictly adhere to least-privilege principles. Without continuous oversight, automated agents risk becoming exposed targets for prompt injection attacks, unauthorized privilege escalation, and unintended system modifications.
Ultimately, enterprise technology leaders must balance rapid automation with proactive security governance. Implementing robust identity verification, continuous behavioral auditing, and strict privilege controls ensures that the ongoing expansion of agentic AI infrastructure delivers high operational efficiency without compromising core enterprise security.