Are UAE enterprises ready for agentic AI?

Security as a critical consideration

Like all new technology implementations, agentic AI operations bring along their own set of its own vulnerabilities.

“There is a growing risk associated with the rise of agentic AI, autonomous agents communicating with other agents and applications without direct human oversight,” said Tony Zabaneh, director of systems engineering for Middle East South at Fortinet. “Without visibility into the Model Context Protocol (MCP), the mechanism that enables AI agents to share context and coordinate actions, these interactions occur outside traditional controls.

“This creates visibility gaps that can expose sensitive data or enable actions to be taken without oversight. Not only that, as AI agents start performing human-like tasks, they may need elevated permissions across multiple applications, which can create security risks at integration points, potentially leaving the door open for attackers to gain control of a bot system that could lead to complete access to organisational operations,” he added.

New risks, such as direct or indirect prompt injection attacks, a technique used to manipulate an AI system by inserting instructions that override, bypass or alter the model’s intended behaviour, can result in unauthorised accessibility to data sources. This technique is the AI equivalent of social engineering in online operations. Instead of hacking the software directly, an attacker tries to trick the AI into following malicious or unintended instructions hidden in documents, emails, websites or data sources to manipulate AI behaviour. 

“In terms of identity management, organisations should look into implementing solid zero-trust principles, the principle of least privilege (PoLP), hardening credential management with tools including Privileged Access Management (PAM), establishing continuous monitoring and anomaly detection, network segmentation and improving visibility across the whole system,” said Zabaneh.

Check Point’s 2026 Cloud security report found that 77% of organisations have updated their security strategy in response to AI, yet only 26% say they have the architecture in place to enforce it consistently. At the same time, 64% have already deployed AI agents in pilot or production environments, underscoring that autonomous AI is advancing faster than the controls designed to govern it. 

“Enterprises need a prevention-first governance model embedded into the architecture itself,” said Ram Narayanan, country manager at Check Point Software Technologies for the Middle East. “That means unified visibility across users, applications, data and agents; real-time inspection of prompts, responses, and tool use; and runtime controls that can enforce policy before risky actions are executed.”

Strong governance is equally important. “The guardrail conversation has to come before anything else. The relevant departments must understand what data an agent can see, and what it can trigger, and the practical version of the exercise is to keep those boundaries tight and monitor agent behaviour continuously,” explained Hetarth Patel, vice president MEA, Americas & Asia Pacific at WebEngage. 

“Another critical safeguard is grounding AI in trusted enterprise data and underlying systems of record,” Easwar Arumugam, product head, Chia, Ramco Systems, added. “By ensuring that all actions are executed through verified data sources and APIs, companies reduce exposure to manipulation and hallucination-driven errors.”

Towards enabling efficient governance

The UAE's two-year Agentic AI push will accelerate modernisation across the private sector. Faster licensing, regulatory approvals, procurement and citizen services will encourage businesses to redesign operations around real-time digital interactions rather than traditional bureaucratic processes.

Interestingly, Trends Research and Advisory’s recently released analytical report, The algorithmic state: How AI and machine intelligence are reshaping the future of government, examines the transformative role of AI and machine-to-machine (M2M) communication systems in modern governance. The report identifies three transformative governance patterns enabled by AI, namely decentralisation, adaptive institutions and hybrid governance.

Decentralisation enhances resilience and scalability by allocating decision-making across algorithmic units and M2M networks. Adaptive institutions rely on real-time information and continuous feedback to dynamically revise policies, considerably reducing the traditional review processing time. Finally, hybrid governance ensures that ethical oversight, contextual interpretation and public accountability remain central to decision-making through human-in-the-loop and human-on-the-loop models as standard governance approaches.

In this new operating model transformation, enterprise operations will have to adapt their processes, governance, talent and technology fast enough to keep pace with a transformation already underway.

To succeed in the Agentic AI era, experts identify four critical priorities: developing AI-native talent and culture, building modular API-first architectures, embedding governance by design, and redesigning operating models for real-time orchestration and decision-making.

Organisations will have to adopt controlled autonomy, automating low-risk, high-volume tasks while maintaining human oversight for decisions with significant financial, legal, regulatory or reputational implications in line with the Federal Authority for Artificial Intelligence and Data guidelines.