Malik Haidar is a distinguished cybersecurity authority who has spent years navigating the complex intersection of threat intelligence and corporate risk management for multinational enterprises. With a career rooted in the strategic integration of business logic and security operations, he offers a unique perspective on the rapid evolution of autonomous technologies. As the industry experiences a monumental shift toward agentic systems, Haidar’s expertise provides a vital roadmap for understanding how these self-executing tools will reshape the defensive perimeter.
The following discussion explores the transformative wave of mergers and acquisitions in early 2026 that have prioritized AI and agentic systems. We delve into the creation of new security categories, the rise of AI gateways to manage privileged digital insiders, and the architectural shifts required to secure data lakes and identity layers as autonomous agents become the primary drivers of enterprise workflows.
Palo Alto Networks recently pioneered a category called Agentic Endpoint Security to address vibe coding and autonomous endpoint tools. How does this shift redefine traditional endpoint protection, and what specific steps should security teams take to defend these autonomous tools against both external exploits and internal logic failures?
The shift toward Agentic Endpoint Security, or AES, represents a fundamental move away from reactive posture management toward the real-time defense of self-coding systems. When Palo Alto Networks acquired Koi for over 0 million, they recognized that “vibe coding” and autonomous tools create a fluid environment where the code itself is constantly changing without human oversight. To defend these, security teams must implement monitoring that can distinguish between a legitimate autonomous optimization and a logic failure that could be exploited by an adversary. We are seeing a move where protection must live alongside the innovation, ensuring that as agents act on the endpoint, their “intent” is validated against organizational security policies. It is no longer enough to scan for known malware; we must now secure the very process of autonomous creation at the edge.
The emergence of AI Gateways aims to balance developer flexibility with centralized control for autonomous agents acting as privileged insiders. What technical hurdles do organizations face when integrating these control planes, and how can you measure the effectiveness of security protocols that govern high-speed agentic decision-making?
The primary technical hurdle is latency; you cannot have a security gateway that slows down the decision-making speed of an AI agent, or you lose the business value of the automation. By acquiring Portkey to serve as an AI Gateway for the Prisma AIRS platform, the goal is to create a foundational layer that treats agents as highly privileged insiders who require constant, transparent oversight. To measure effectiveness, organizations should track the “intervention rate”—how often the gateway must block or modify an agent’s request—and the “alignment score” between agent actions and security guardrails. It is a delicate balance because developers need the flexibility to scale AI in production, but security teams need a kill-switch that can operate at machine speed. This central control plane becomes the only way to manage agents that have the authority to move data and change configurations across the enterprise.
Creating data lake solutions specifically for AI agents highlights a growing need for instantly traceable and secure data access. What are the primary risks of using standard data storage for agentic AI, and what architectural changes are required to ensure data remains secure without sacrificing the speed of automated workflows?
Standard data storage often lacks the granular, high-speed telemetry needed to track how an autonomous agent interacts with specific datasets, which creates a massive visibility gap. When Cyera acquired Ryft, they addressed the need for “instantly traceable” data access, which is something traditional lakes simply weren’t built for. The risk of using legacy systems is that an agent could inadvertently leak sensitive information or ingest poisoned data, and you wouldn’t know until the damage was done. Architecturally, we need to move toward a unified control plane where data security is baked into the retrieval process itself. This means building environments where every data point accessed by an agent is tagged and logged in real-time, allowing for a shared vision of security that supports the speed of agentic workflows without compromising the integrity of the information.
Moving toward autonomous runtime identity security requires AI-driven access control that matches the speed of the agentic era. How do these real-time systems differ from traditional identity management, and what metrics should security leaders track to ensure these agents do not inadvertently create new vulnerabilities during automated sessions?
Traditional identity management is often static and based on human-centric login events, whereas autonomous runtime identity security must handle thousands of micro-transactions per second. Silverfort’s acquisition of Fabrix is a perfect example of this evolution, aiming to create a system that provides access control at the exact moment an agent attempts a task. Security leaders need to track metrics like “privilege duration”—ensuring that an agent only has elevated rights for the millisecond it needs them—and “identity velocity,” which monitors for abnormal spikes in access requests. This next generation of identity security is designed to stand up to AI-driven adversaries who exploit the gaps in slow, manual approval processes. By automating the identity layer, we can ensure that these agents don’t become the weakest link in the corporate perimeter.
Integrating privacy-first identity stacks with agentic workflows is becoming a priority for both specialized tech firms and broad HR platforms. What are the practical trade-offs when implementing “proof-of-human” verification alongside automated systems, and how can companies maintain a secure identity layer as AI takes over more operational functions?
The trade-off often lies between user friction and absolute certainty; implementing “proof-of-human” can slow down a workflow, but it is the only way to ensure a malicious bot isn’t masquerading as an employee. Self Labs’ acquisition of Loam highlights the urgency of building a “trust layer” that maintains user privacy while allowing multi-agent systems to function. Even HR platforms like Remote are getting involved, acquiring Bravas to unify identity and device management because they realize that as AI takes over payroll and HR functions, the identity layer becomes the ultimate safeguard. To maintain security, companies must make IT “invisible” by automating the verification process so it happens in the background. This ensures that even as agents handle complex operational tasks, there is a verified human or a strictly governed digital identity at the root of every action.
What is your forecast for Agentic AI security?
I forecast that by the end of 2026, the concept of “manual” security configuration will be viewed as an obsolete relic of the past. We will see the total convergence of identity, data, and endpoint security into a single, autonomous defensive fabric that mirrors the speed and complexity of the agents it protects. The acquisitions we saw in April—from Palo Alto Networks to Silverfort—are just the first wave of a massive consolidation where every major vendor must own an “agentic” security stack to survive. Ultimately, the winners in this market will be those who can provide a “proof-of-trust” for every automated action, turning cybersecurity from a bottleneck into the primary enabler of the autonomous enterprise. Security will no longer be a separate department but a silent, high-speed operating system that governs every digital interaction.
