At MEBIS 2026, banking leaders focused on the practical challenges of putting increasingly capable AI into real financial workflows. 

[For more news, click here]

Generative AI has moved quickly through banks, from customer service tools and employee assistants to applications that help with fraud, compliance and risk management. At the Middle East Banking Innovation Summit in Dubai this month, much of the discussion moved to the next step: systems that can do more than generate an answer and instead retrieve information, interact with other applications and carry out parts of a workflow. 

MEBIS 2026, held September 16 and 17 at Jumeirah Emirates Towers, brought more than 400 banking and technology leaders together for the summit's 17th edition. Agentic AI featured prominently in the program, alongside data, cybersecurity, cloud, automation and payments. A discussion titled "From AI Assistants to AI Agents: Autonomous Banking, Governance and Trust" examined what happens when generative AI moves from assisting employees to executing parts of a workflow. The event's official agenda also included sessions on AI and intelligent automation, building an AI-ready bank and the infrastructure required for wider AI adoption. 

When AI Starts Taking Actions 

The distinction matters because banks are already using AI in areas where mistakes can have financial and regulatory consequences. The Bank for International Settlements has reported that financial institutions are deploying AI in areas including fraud detection, credit assessment, compliance, customer service and risk management. Its recent work has also examined potential uses for AI agents in financial market infrastructure. The next step involves extending existing AI capabilities into processes where software can initiate actions rather than simply provide information. 

A customer-service assistant that drafts a response can be reviewed before anything happens. An agent connected to customer data, internal applications or payment systems may be able to perform the next step without that review. Banks then need to establish what the system is allowed to access, which actions require approval and how its decisions and activity can be traced afterward. 

Security is becoming part of that discussion. An AI agent can have legitimate credentials and access to an enterprise environment while still carrying out an action beyond the task it was intended to perform. Controlling those systems therefore involves more than deciding who or what can log in. Banks also need controls around what an agent can do after it has gained access, with the ability to restrict, stop or escalate actions when necessary. 

The Technology Behind the Agent 

The AI model is only one part of the deployment problem. A bank's agent may need to draw information from customer databases, identity systems, core banking platforms and other applications that were built at different points in the institution's technology history. The reliability of those connections, the quality of the underlying data and the controls around access can determine whether an agent is useful in practice. 

MEBIS reflected that reality by placing data, cloud architecture, cybersecurity and intelligent automation alongside its discussions of generative and agentic AI. If software is going to act on behalf of a bank, the institution has to know what information it is using, what authority it has and what happens when it encounters something outside the instructions it was given. 

Human oversight remains part of that equation. Some banking processes are repetitive enough to be handled largely by software, while others involve exceptions or decisions where context matters. Employees can spend less time checking routine work as agents take on more tasks, but responsibility for decisions and exceptions still has to remain clearly assigned within the institution. 

A Broader AI Conversation 

MEBIS placed agentic AI within a wider discussion about how regional banks are changing their technology operations. The program covered digital banking, customer experience, payments, Open Finance, risk and fraud alongside AI, putting the technology in the context of the systems it will have to work with rather than treating it as a separate project. 

“What stood out at MEBIS this year was how quickly the conversation around AI has matured. We are no longer talking about AI as one technology initiative among many. It is beginning to influence almost every part of the bank, from how customers are served to how decisions are made, risks are managed and operations are run. The quality of the discussions at MEBIS 2026 showed that banks across the region are not simply watching this transformation; they are actively preparing for it,” said Shail Bisht, Regional Director at Expotrade Middle East, 

The more immediate issue for banks is where AI agents should be given authority to act without intervention and where a person should remain in the loop. Decisions around data, identity, integration and security will matter just as much as the models themselves. MEBIS 2026 showed the regional banking conversation moving into that practical stage, with attention turning from what AI can produce to the conditions under which banks can put those capabilities to work. 

Related Articles 

Why Human Oversight Is Becoming the Real Scaling Engine for Gulf Fintech's AI Agents 

Delinea Takes AI Agent Security Beyond Access 

The AI Agent Security Gap Is Closing, and the Middle East Is Proving to Be the Fastest Market to Demand It