An Abu Dhabi Asset Manager Just Wrote ADGM’s Rules Into Code. The FSRA Did Not Sign the Code.
An Abu Dhabi asset manager embeds ADGM regulations into smart contract code, raising questions about regulatory oversight, compliance accountability, and governance.
On 27 October, Further Asset Management, an Abu Dhabi asset manager based on Al Maryah Island, said it had embedded Chainlink’s Automated Compliance Engine in Anubis, a system it describes as built for machine-readable regulation. The engine checks tokenized asset transactions against rule sets covering FSRA requirements for issuance, KYC, AML, ownership, trading, custody and sanctions, and it does so before the transaction executes. Initial deployment is in ADGM. A dashboard for virtual asset service providers and regulators is planned.
Coverage reads it as ADGM automating compliance. The release is more careful than that.
Every operative verb is hedged. The workflows are “designed to support compliance with applicable” FSRA requirements. The controls “aim to comply with the transaction rules that are prescribed by ADGM and the FSRA.” Nowhere does the release say the regulator wrote, reviewed, or recognizes the rule set. There is no FSRA quote. No adopting institution is named; the parties “plan to announce further milestones as the first wave of institutions adopts Anubis.”
So what launched is one firm’s reading of the FSRA rulebook, executed at machine speed. That is useful. It is not a transfer of responsibility, and the difference is the whole story.
The rules moved off the token
Chainlink’s own description of ACE explains where the control now sits. Compliance rules are configured by the asset issuer through a policy manager that runs offchain, inside Chainlink’s runtime environment, and the result is enforced onchain. Chainlink Labs’ Johann Eid put the benefit plainly in the release: compliance teams can “update those rules in real time without having to redeploy token contracts.”
For anyone who has watched a token contract need a fresh audit because a jurisdiction changed one threshold, that is a real improvement. It also changes what a token is. Under the older model, transfer restrictions were written into the contract, audited once, and visible to anyone who read the code. Under this one, the token asks an external policy what it is allowed to do, and the policy can change between one transfer and the next.
The control that matters most is now a configuration. Someone has to own it, version it, approve each change, and be able to show a supervisor which rule set applied to a given transfer on a given day. None of that is a Chainlink question or a Further question. It belongs to whoever is licensed for the activity.
The license does not move
ADGM’s General Rulebook is direct on this. GEN 3.3.31 says an Authorized Person that outsources functions directly related to its Regulated Activities, including to a member of its own group, “is not relieved of its regulatory obligations and remains responsible for compliance with the Regulations and Rules.” The outsourced function “shall be deemed as being carried out by the Authorized Person itself.” The firm must supervise the provider effectively and deal with any failure that leads, or might lead, to a breach.
Read against that rule, a fund manager that routes investor eligibility through Anubis has not handed off a compliance obligation. It has added a service provider it must supervise. If a transfer clears the engine and later turns out to breach an FSRA rule, the FSRA’s conversation is with the manager. “The engine approved it” is a description of the failure, not a defense.
That is not a criticism of Further or Chainlink. It is how outsourcing works in every financial center that borrowed its rulebook from London, and ADGM is one of them.
What each seat gets
Further’s motive is the most interesting. It is an asset manager, and Anubis gives it a product that reaches beyond its own funds: if other ADGM managers and venues run their tokenized transfers through its rule sets, Further becomes the compliance layer for a market of 171 asset and fund managers and more than 244 funds, by the release’s own count. That is a better business than managing one more fund.
Chainlink gets a Gulf reference client for ACE inside a regulator that has been running a virtual asset framework since 2018.
Adopting firms get cheaper pre-trade checks and a new dependency. The awkward seat is the competing manager. Its investor eligibility logic, which defines who its product is for, would run through a rule set maintained by another asset manager. Some will accept that for the cost savings. Many fund managers will want to see the rules, control the changes, and keep the audit trail on their own side. That tension decides who adopts first.
Where an engine works, and where it cannot
The real question is which compliance checks are binary and which are judgments.
Binary checks encode well. Is this wallet linked to a verified legal entity? Is the counterparty on a sanctions list? Has KYC expired? An engine answers these faster and more consistently than a person, and Chainlink’s identity layer is built for them.
Judgments do not encode. Take the gate: every Exempt Fund in ADGM depends on Professional Client status. Under COBS 2.4.4, an individual can be an “assessed” Professional Client with net assets of at least US$1 million (excluding the primary residence) and if the individual “appears, on reasonable grounds, to have sufficient experience and understanding of relevant financial markets, products or transactions and any associated risks.” The firm must hold objective evidence of the assets and their valuation. An engine can check that a credential saying “Professional Client” exists. It cannot make the assessment the credential records, and the firm that made it carries the risk if it was wrong.
Shariah screening divides the same way. In ADGM, a fund that holds itself out as Islamic answers to its own Shariah Supervisory Board under the FSRA’s Islamic Finance Rules. Some of what that board decides is binary and travels well into code: a list of excluded sectors, an exclusion of interest-bearing instruments. Some depend on data that moves, such as the share of tangible assets in a sukuk portfolio, which changes over an instrument’s life and needs a live feed from an issuer the fund has no contract with. And some cannot be configured at all until a person translates it: a board’s ruling on a structure it has not seen before. An engine can enforce the screen. It cannot hold the ruling, and a rule change made “in real time” has to match a ruling the board actually gave.
So the engine is low-risk where the rule is a fact and high-risk where the rule is an opinion. Most of what makes a tokenized fund distribution-ready sits on the opinion side.
The call
None of this makes Anubis less useful. It makes the workaround more specific. A manager adopting a pre-trade engine still has to decide which checks go into it and which stay with people, who signs off a change to the rule set, how a Shariah board’s rulings are mapped into configuration and kept in step, and how the outsourcing file under GEN 3.3.31 will read when the FSRA asks for it. Zamanat does that mapping for managers bringing Shariah-aligned products onchain in ADGM and DIFC, on the manager’s side of the table, where the license sits.
Two calls, each with a test. First, the first institution named as an Anubis adopter will be a tokenization venue or a virtual asset service provider, not a fund manager, and it will be named within six months. If a third-party ADGM fund manager is named first, the call is wrong. Second, the FSRA will not publish any statement recognizing rule-engine output as evidence of compliance within twelve months. Adopters will keep running parallel human checks on the judgment-based rules, and the cost savings will be smaller than the release implies.
The planned regulator dashboard is the piece to watch. A regulator that looks at it is reading Further’s interpretation of its own rules. Whether the FSRA treats that view as a convenience or as a record will settle how far this goes.