The agentic sukuk lifecycle: a reference design with five human gates

Three principles before the stages

1. The agent executes; it never rules. Scholars write the rules and decide the grey cases. The agent applies the rules to facts and surfaces the grey cases. This follows directly from rulings such as Dar al-Ifta's fatwa 22255 on AI and fatwas, and from the structure of every Shariah governance framework we know of.

2. Every action is a logged tool call. The agent acts only through named tools (a registry, a payment rail, a screening service), through a standard interface such as the Model Context Protocol, with input, output and rule version recorded. If a step cannot be logged this way, it is not automated.

3. Uncertainty escalates; it is never estimated. When the facts do not clearly match a rule, the agent stops that thread and hands it to the relevant gate with the evidence attached. "Probably fine" is not an output the agent is allowed to produce.

Stage 1: Structuring

Agent work. From a template the Shariah board has approved for this structure type (say, ijarah with a service agency), the agent drafts the term sheet, populates the asset schedule from the originator's records, and runs the pre-issuance tests:

• Is each asset tangible, identifiable and legally transferable?
• Is the asset's current use on the exclusion list? Is the tenant's business activity?
• What share of the pool is tangible versus receivable, against the board's minimum?
• Is the purchase undertaking priced in a way the board has previously approved for this structure?
• Are the documents consistent with each other (dates, amounts, parties)?

Output: a checklist with a pass, fail or unclear mark on each line and a link to the evidence.

Gate 1: Shariah board approval. Named scholars read the checklist and the documents and approve, reject or send back. Their decision, the rule version they applied and the date are recorded. The agent cannot proceed without this record.

What could go wrong. The template itself is flawed. Mitigation: templates are board-approved artefacts with version numbers and review dates, and a change to a template is itself a Gate 1 event.

Stage 2: Issuance

Agent work. Investor onboarding: identity verification, eligibility against the offering's rules (institutional only, jurisdiction limits), suitability flags. Allocation of certificates according to the book. Settlement: cash received against tokens or certificates delivered, reconciled line by line.

Gate 2: Regulator and issuer sign. The regulator's approval of the offering and the issuer's signature on the documents are human acts and stay that way. The agent's role is to make sure that when the humans sign, the facts in front of them are complete and reconciled.

What could go wrong. Mis-allocation, or onboarding someone who should have been excluded. Mitigation: the reconciliation output is reviewed before settlement is released, and any eligibility "unclear" blocks that investor rather than admitting them.

Stage 3: Servicing

Agent work. Each period: confirm the rent or profit received from the originator; compute each holder's share; execute distributions; reconcile the ledger; prepare the periodic report. Flag any shortfall immediately rather than at the next report.

This is where the money is. Servicing a multi-year sukuk with thousands of holders is labour-intensive, error-prone and entirely rule-based. On a tokenised sukuk, distribution can execute as a transaction; on a conventional one, the agent prepares instructions a human releases.

Gate 3: Auditor attestation. An external auditor reviews the logs and the accounts at agreed intervals. Because every distribution is a logged tool call with its inputs, the audit is a replay rather than a reconstruction.

What could go wrong. A wrong input (for example, a rent figure entered incorrectly by the originator) propagated to every holder. Mitigation: two-source confirmation of inbound cash, and a tolerance check that halts distribution if the figure deviates from the schedule.

Stage 4: Monitoring

Agent work. Continuous, not annual. The agent watches:

• Asset use. Tenant activity codes, licences, public records and news for signs that a leased asset is being used impermissibly.
• Pool composition. The tangible-asset share, measured daily, against the board's minimum, with an early warning before the line is crossed.
• Originator events. Missed payments, rating changes, insolvency filings, material corporate actions.
• Covenant compliance. Insurance in force, maintenance obligations met.

Output: flags with evidence, ranked by urgency, routed to the right gate.

Gate 4: Scholars rule on breaches. When a flag concerns Shariah compliance (a tenant started selling alcohol; the pool fell below the minimum), named scholars decide what it means and what must happen: cure period, purification of tainted income, suspension of trading, or no action. Legal and credit flags route to the trustee and the issuer under the documents.

What could go wrong. False positives that exhaust the scholars, or false negatives that miss a real breach. Mitigation: the agent's detection rules are themselves board-approved and reviewed on a schedule, with precision and recall reported.

Stage 5: Maturity

Agent work. On the scheduled date, or on an early redemption event: compute the final distribution; execute the purchase undertaking at the documented price; transfer the asset back to the originator; return capital to holders; file the closing report; archive the complete log.

Gate 5: Trustee confirmation. The trustee confirms that the asset transfer and capital return occurred as documented and releases the parties. On a tokenised sukuk this is a verification of ledger state; on a conventional one, of registrar and bank records.

What could go wrong. A dispute about whether the undertaking is enforceable, as in Dana Gas. The agent documents; courts decide. No design removes legal risk, and a design that claimed to would be lying.

The log: what gets recorded at every step

An auditor, a Shariah board or a regulator should be able to answer "why did the agent do X on date Y" from this table alone.

How to spot a dangerous design

• A gate is missing. If a product says the agent "approves" structures or "resolves" breaches, a human decision has been removed. Ask which rule authorised that.
• Rules without versions or review dates. A 2026 ruling encoded forever.
• No replayable log. "Our AI handles it" is not an audit trail.
• Estimation under uncertainty. Any output like "likely compliant" from an agent is an agent making a ruling.
• No named person. Accountability assigned to a vendor or a model rather than a person at the institution.

Where the market is on this ladder

Nowhere yet, honestly. The most advanced live infrastructure is the tokenised layer described in tokenised sukuk, which the April 2026 Khazanah pilot in Malaysia established. Agents at the "assisted" level, drafting and checking with human approval of each step, exist as internal tools in banks. Nobody has published an agentic sukuk that runs servicing or monitoring with the authority described here. The home page tracks the ladder and moves the marker when evidence arrives.

Using this design

Anyone may use this reference design, with attribution, to evaluate or build a product. If you are building, send us the design and we will list it here with a comparison against the five gates. If you are a scholar, regulator or auditor and you think a gate is in the wrong place, tell us and we will publish the correction with your reasoning. The contact address is on the about page.