Agentic AI has become the organising theme of banking technology. The shift it describes is precise. Software no longer proposes an action for a person to approve. It takes the action itself, at machine speed, across thousands of accounts at once.
That shift puts one question in front of every bank. When a system acts on its own, who can explain why it acted that way?
For Islamic banks, the question arrives twice.
Most banks cannot answer it yet
SaaScada research among 150 UK banking innovation leaders, conducted in March 2026 across institutions with balance sheets from £0.5bn to £100bn, found near-universal belief in agentic AI and very little readiness for it.
91% believe agentic AI will enable new ways of designing banking services. 31% actively deploy AI in core operational or decision-making processes.
The gap sits in the evidence. The SaaScada AI study found only 12% are very confident they could explain an AI-driven decision to their regulator. 77% say legacy systems restrict data availability. 71% struggle to access real-time data at all.
The study covered UK institutions, and what it describes is not a UK condition. It is what happens anywhere ambition meets a core built to report on the day that has already finished.
Islamic banks answer to two arbiters
Every bank answers to a financial regulator. An Islamic bank answers to a Sharia board as well.
The Internal Sharia Supervision Committee is a body of qualified scholars appointed by the bank’s own board. It reviews and approves every product, contract template and policy before launch. It rules on changes. It issues fatwas that bind the institution. At year end it writes a report stating how far the bank complied during the year.
In the UAE that report goes to the Higher Sharia Authority, a regulatory body established within the central bank under Decretal Federal Law No. 14 of 2018, which reviews and approves it before the bank may present it to shareholders. The report is then published within the financial statements. Malaysia, Bahrain, Pakistan and Indonesia run comparable structures.
The practical consequence is that an Islamic bank cannot self-certify. Product approval, ongoing control, independent Sharia audit and external regulatory sign-off are four separate steps, each requiring evidence. All four now have to govern systems that act faster and more autonomously than most controls and committees were designed to oversee, so complete data that aids explainability is paramount.
None of this is new pressure on Islamic banks, and conventional banks carry no less governance for it. What differs is the second arbiter, which has asked for evidence of how decisions were reached since long before AI. That makes agentic AI in Islamic banking a harder test to pass, and a better guide to what the rest of the industry will soon need.
What technology can and cannot do about Sharia compliance
Sharia compliance is certified at the level of products and processes. A scholar rules on a contract structure, on how a profit share is calculated, on how a late payment is treated, on whether a Murabaha mark-up was disclosed as the contract requires. Certification attaches to those things.
No platform holds that certification. No vendor can hand it to a bank. Compliance belongs to the institution and its scholars. What technology decides is narrower. It determines whether the bank can evidence its own compliance, and how quickly it can act on what its Sharia board rules. Those two things are squarely an architecture question, and they are where agentic AI raises the stakes.
An agent that adjusts a customer’s profit-share allocation, flags an account for restructuring, or prices a diminishing Musharaka instalment is doing work the Sharia board approved in principle. Demonstrating that it stayed inside that approval, on any given account, on any given day, becomes a data challenge.
Four questions worth asking of your core
For an Islamic bank or the Islamic window of a conventional bank assessing whether its foundations can carry agentic AI.
1. Can it show how any figure was reached?
Not what the balance is today, but the sequence of events that produced it. When a Sharia auditor asks how a profit distribution was calculated eight months ago, reconstructing an answer from month-end snapshots is not the same as showing the record. A core that stores every event as it happens has the answer already. A core that stores balances has to rebuild one and hope it matches.
2. Is the data available now, rather than later?
Agents act in the moment. An agent working from a batch is acting on a version of the customer that no longer exists. Real-time data is what makes the action appropriate, and the same real-time stream is what lets the bank see what the agent did while it still matters.
3. Is the separation of funds provable in the ledger?
Where a bank runs Islamic and conventional business together, through an Islamic window or a separate brand, the two pools must not commingle. Separation that exists only in the reporting layer is a reconciliation exercise waiting to be questioned. Separation held at entity level in the data itself is simply a fact about the system.
4. Can the product change as fast as the ruling?
Sharia boards rule, and products then have to reflect the ruling. When a change takes a development cycle, the bank pays twice, once in build cost and once in the lag between decision and implementation. When a product team can configure the change itself, with every version recorded and the timing under its control, the approval cycle stops being the constraint.
These are the properties a modern core banking platform should be judged on for this work: a full-fidelity event-level record, real-time data immediately accessible to the bank, ledger-level product separation, and product change that does not queue behind engineering.
The instinctive bank
Writing about AI and UK banking earlier this year, SaaScada co-founder and president Steve Round set out the test plainly. Banks, he wrote, must be able to answer: “why was this decision made, and can you prove how the model arrived at it?” Without trusted, timely data and systems that can provide it, he argued, that is impossible.
He gave the alternative a name. “Let’s call it instinctive banking.” Instinctive organisations break down internal silos and embrace external partnerships. They harness data and advanced technologies to gain real-time actionable insight, accurate prediction and explainability, so they can anticipate and safely act on customer needs, business opportunities and risks.
Applied to an Islamic bank, instinctive has a particular meaning. It is anticipation exercised inside the policy the bank and its scholars have set, with the evidence of that restraint available on demand. A bank that can show its workings is a bank that can afford to let software act on its behalf. One that cannot, will keep agentic AI where much of the industry currently keeps it, in pilots that demonstrate well and never scale.
The opportunity is real. Islamic banking operates in 84 markets and sits inside an industry that reached US$5.98 trillion in assets in 2024, projected to reach US$9.7 trillion by 2029 (ICD-LSEG Islamic Finance Development Report 2025). The institutions that serve it are not short of ambition. What decides who moves fastest and safest is whether the foundations can carry the weight.
Talk to us about agentic AI in Islamic banking at Seamless Middle East
Steve Round is speaking at Seamless Middle East, 22 to 24 September at Dubai World Trade Centre, on Core Banking of the Future: Real-Time Processing Driving Operational Efficiency. This session, at 14:30 on Tuesday 22 September on Stage 1 of the Fintech Forum, will cover what it takes to modernise and use agentic AI safely for efficiency gains, speed to market and improved customer experience. The SaaScada team is on stand G66 in the Core banking area of Sheikh Saeed Hall 2, across all three days.
If you are weighing up what agentic AI would ask of your core under Sharia governance, or what it would take to build Islamic products on the same platform as the rest of your range, come and put the four questions above to us directly. And also ask if you can achieve this for conventional and Islamic banking products without having to buy separate modules.
