Where the obligations fall
Every framework named here attaches its duties to a kind of entity: a licensed issuer, a registered service provider, an obliged institution. Tesoro is a library. It is none of those things, it cannot become one, and this page exists so nobody has to guess which parts of that are being claimed.
Why state this at all, rather than say nothing? Because silence in this area reads as a claim. A governance library that lists regulations without saying who they bind invites the reading that installing it addresses them — and that reading is wrong in a way that only becomes visible during an examination. Saying plainly what this is not is both safer for a reader and more useful than a page of logos.
Who each one binds
| Framework | Binds | Where this sits |
|---|---|---|
| GENIUS Act (US) | Permitted payment stablecoin issuers — federal or state-qualified entities | Outside. A library is not an issuer and cannot be permitted |
| MiCA (EU) | EMT and ART issuers, and crypto-asset service providers | Outside. Not an issuer, not a CASP |
| FATF R.16 (Travel Rule) | Ordering and beneficiary institutions; VASPs for virtual-asset transfers | Outside. It defines a record and never a message — see below |
| FCA (UK) · MAS (SG) · HKMA (HK) · VARA (AE) | Licensed and registered firms in each jurisdiction | Outside. Licensing is organisational; software is not licensable |
The entity carries the obligation, always. If you are one of those entities and an agent spends through you, the exposure is yours — and it arrives without a category, because an autonomous agent is not an obliged entity under any of the seven. That is correct rather than an oversight: an agent has no legal personality and cannot hold a licence.
The GENIUS timeline, with its dates
| When | What |
|---|---|
| 18 July 2025 | Signed into law |
| by 18 July 2026 | Federal regulators are to issue implementing regulations through notice-and-comment rulemaking |
| 8 April 2026 | FinCEN and OFAC issue a joint proposed rule for permitted payment stablecoin issuers — AML/CFT programmes, sanctions compliance programmes, and the technical capability to block, freeze and reject specific transactions. Published in the Federal Register on 10 April 2026; comments closed 9 June 2026; proposed to take effect 12 months after final issuance |
| by 18 January 2027 | The Act takes effect — the earlier of 18 months after enactment or 120 days after final implementing regulations. It is a ceiling, not a fixed date |
What was read, and what was not. The dates above come from the Federal Register listing and from published law-firm analyses, cross-checked against each other on 28 August 2026. The NPRM itself was not read in full. The FATF material elsewhere on this site was read in the source text — the consolidated Recommendations 2012–2025 and the June 2025 explanatory note — and that difference is stated rather than smoothed over.
One word changed in June 2025, and it moved away from agents
The FATF glossary defines originator. The June 2025 revision changed it, and the change is one word:
“Originator refers to the account holder who allows requests the
payment or value transfer from that account…”
“Allows” would have covered an account holder who authorised a standing policy. “Requests” does not. The definition narrowed to an act of requesting, at the moment software began doing the requesting.
Two things to be careful about, because this is the kind of finding that gets overstated.
| Careful about | What is actually true |
|---|---|
| It is a consequence, not an intent | The revision's stated purpose is payment transparency and ISO 20022 alignment. The phrase artificial intelligence appears nowhere in the explanatory note, and “requests” is arguably the more natural description of what an account holder does. Nothing suggests agents were considered. |
| Fiat rails do carry an initiator field | CBPR+ pacs.008.001.08 has InitiatingParty [0..1] and
UltimateDebtor [0..1]. So “no payment format can express this” is
false, and an earlier version of this analysis said it. It is optional, absent
from R.16's required set, names a party rather than a mode, cannot name an agent
— an agent has no legal personality — and yields the same value whether authority
was delegated or credentials were stolen. IVMS101, the virtual-asset payload, has no
equivalent at all. |
What survives is narrow and checkable. R.16's required set — name, account number or unique transaction reference, address or national identity number or customer identification number or date and place of birth, plus LEI or BIC for a legal person — identifies whose money moved and never who instructed the movement. An agent-initiated payment therefore produces a transfer record naming a person who pressed nothing, indistinguishable from one a person reviewed and sent and from one made with stolen credentials.
Check it rather than take it. Both documents are free from
fatf-gafi.org: the consolidated FATF Recommendations 2012–2025 and the
Explanatory note for revised R.16. The glossary entry is in the consolidated text; the
struck word is visible in the explanatory note's redline. Retrieved and read
28 August 2026. Full working, including the searches that returned zero hits for
initiating party and artificial intelligence, is in
the
verification record.
One line in the April proposal is worth reading carefully
The FinCEN and OFAC proposal asks permitted issuers for the technical capability to block, freeze and reject specific transactions. That is a requirement about mechanism, not only about policy, and it points the same way this project does: toward controls that act before a transfer settles rather than reports that describe it afterwards.
It is still not a requirement this software satisfies. The obligation is on the
issuer. What a library can offer is a place for that capability to be exercised and evidenced:
Governor.freeze() refuses everything and says why, persisted, attributed to
killswitch, and an unreadable state file reads as frozen rather than as
permission. The sanctions control is the highest-precedence dispositive control, so a
real screening provider has somewhere to land — and no engine sits behind it
today.
What is not claimed, in plain terms
Not compliance
With GENIUS, MiCA, the Travel Rule, or any framework named on this site. None is claimed or sought, and no configuration of this software produces it.
Not AML screening
No list, no matching, no jurisdiction model. The stablecoin-1 profile requires an
AML position in the record and this implementation does not conform to it.
Not Travel Rule support
It defines a record, never a transfer. Transmitting those fields, or presenting their presence as satisfying an obligation, is forbidden by the standard itself.
Not audited
Every conformance case and adversarial attack was written by the author of the system under test. That is a real limitation and it is not fixable by more of the same work.
What an obliged entity actually gets
Evidence, not compliance — and the distinction is the whole design. For a spend that an agent proposed, a record that answers: which accountable human authority it acted under, which control permitted or refused it, and how the transfer was initiated — the last being a fact R.16's required data set has no field for, because every element there identifies whose money moved and none identifies who instructed the movement.
Whether that evidence helps discharge any particular obligation is a legal determination requiring a qualified person, a jurisdiction and facts a technical layer does not hold. This software does not make that determination and neither does this page.
This page is general information and is not legal advice. It states which entities each framework binds and asserts no compliance with any of them. Dates and scope are as described in the sources named above and may have changed; verify against the instruments themselves before relying on anything here.