Thailand's Same-Owner Test: The Stablecoin Rule That Binds Only the Observable

Over the twelve months surveyed in Thailand's newest digital asset consultation, USDT volume on licensed Thai platforms grew faster than any other supervised asset category. The Securities and Exchange Commission noticed. On that observation it built a rule — and the rule asks licensed operators to verify something no intermediary can actually see.

The proposal, open for comment until September 25, 2026, would require licensed operators to accept stablecoin deposits only from wallets owned by the customer, and to release stablecoin withdrawals only to wallets owned by the customer. A daily ceiling of THB 5 million — roughly $140,000 — would sit on top of that test, with named carve-outs for certain operator businesses, Bank of Thailand-authorized entities, and market makers.

If a rule requires an intermediary to prove a fact for which it has no observation mechanism, the rule does not produce compliance. It produces attestations. That distinction is the substance of this consultation paper, and it is being discussed far less than the headline number.

Thailand's digital asset perimeter runs on the Digital Asset Business Act, supervised by the SEC, with anti-money laundering obligations routed through AMLO under the 1999 Anti-Money Laundering Act. Licensed virtual asset service providers — exchanges, brokers, dealers, custodians — sit inside that perimeter. Unhosted wallets, decentralized exchanges, and person-to-person transfers do not.

The same-owner test is designed to operate alongside the Travel Rule, not inside it. The Travel Rule — FATF Recommendation 16, taking effect in Thailand on February 27, 2027 — governs information transmission between institutions. When one VASP moves assets to another on behalf of a customer, both sides exchange originator and beneficiary data. That is an inter-institutional control with a defined counterparty.

The same-owner requirement is orthogonal. It applies even when the counterparty is not an institution at all: an unhosted wallet, a hardware device, a self-custodied address in a browser extension. The Travel Rule has nothing to say about those counterparties, because there is no second institution to speak to. The SEC's proposal fills the gap by shifting the obligation onto the customer's relationship with their own key material.

The framework is not revolutionary. Variants of the same-owner concept appear in Singapore and Hong Kong, and FATF's VASP guidance supplies the underlying template. What distinguishes the Thai proposal is enforcement posture: it pushes the ownership test down to the counterparty wallet itself, rather than stopping at the institutional relationship.

The regulatory logic is legible. Observe growth in a bearer asset. Identify the money-laundering and cyber-fraud exposure that growth implies. Design an ownership test. Attach a value ceiling. Exempt the scenarios where the test would break the market.

Three exemptions are named: transfers within certain licensed operator businesses, transfers involving entities authorized by the Bank of Thailand, and market maker activity. The consultation does not fully specify how those exemptions interact with the same-owner test itself. That omission is not drafting sloppiness. It is the part of the paper most worth reading twice.

The verification problem decomposes into two legs, and they are not equally hard.

The inbound leg asks a narrow question: does this depositing address belong to this customer? There is exactly one cryptographically sound way to answer it. The operator issues a challenge — a random nonce, usually with an expiry — and the customer signs it with the private key controlling the address. The operator recovers the public key from the signature, derives the address, and compares. On the secp256k1 curve that is a single ecrecover call. The arithmetic is cheap, the verification is deterministic, and the failure modes are well understood.

Everything else is theatre. A screenshot of a wallet balance proves nothing. A signed declaration proves only that a customer will sign declarations. Address-clustering heuristics — grouping addresses by shared spending patterns — produce probabilistic attribution, not ownership, and they degrade the moment a user rotates keys or routes through a mixer. Those heuristics belong in an investigator's toolkit. They do not belong in a rule that obliges an operator to make a binary decision at the deposit screen.

So the sound implementation is a signature challenge. What does it cost?

I spent three days in 2021 reverse-engineering the ERC-721A minting implementation — not because I cared about the art, but because I wanted to see where the gas went. The lesson generalizes beyond mint functions: every verification step bolted onto a transaction path surfaces as throughput, latency, or support tickets. When I audited a STARK-based rollup's circuit design in 2025, the binding constraint was never the arithmetic of verification — it was the cost of generating proofs fast enough to keep the chain fed. Verification is a throughput tax, and it compounds.

Scale it. A licensed Thai exchange clearing a few hundred thousand stablecoin transfers a day cannot run an interactive ceremony on each one. It runs the ceremony once per wallet, caches the attestation, and re-runs it on a schedule or on anomaly triggers. That is the only workable design, and it means the control is a periodic re-attestation regime rather than a per-transaction check. The daily ceiling, by contrast, is per-transaction, because it must be summed in real time. The rule therefore imposes two controls at two different temporal granularities — one event-driven, one continuously aggregated — and requires the operator to reconcile them. Reconciliation logic is where implementation bugs live.

The account model absorbs this reasonably well. UTXO chains do not. On a Bitcoin-family chain, an address is not an identity; it is a script that a set of unspent outputs pays to. Ownership means control of the key that can satisfy that script, and a wallet following standard derivation practice uses a fresh address for every change output. A customer depositing from a UTXO wallet may be asked to prove ownership of an address that has already been spent and holds nothing. The proof still works — the key still controls it — but the mental model breaks, and the support burden is real. That is not an edge case. It is how the majority of self-custodied Bitcoin is held.

Now the harder problem. "Same owner" assumes ownership is a binary attribute. It is not. Custody is a spectrum, and the spectrum has no natural cut point.

Consider four configurations a Thai user might plausibly hold.

A single key in a mobile wallet. Clean case. One challenge, one signature, one answer.

A 2-of-3 multisig — the configuration used by most small-fund and treasury setups. One key on a hardware device, one in a corporate vault, one with an advisor. Who signs the ownership challenge? The advisor alone? Any two? All three? The proposal does not say, and the answer determines whether a corporate wallet can receive stablecoins at all.

An MPC-secured address — the architecture used by institutional custodians. There is no single private key to sign with. The signature exists only as the output of a threshold ceremony. It is cryptographically valid and indistinguishable from a single-key signature at the verification layer. But the operator verifying it has learned that someone completed the ceremony. Not who.

A smart contract wallet with delegated signing authority. Session keys, spending policies, automated recovery modules. Under EIP-7702, an externally owned account can delegate execution to contract code. The address is unchanged. The signing authority is now a program, and the user can revoke that delegation whenever they choose.

In each of the last three configurations, the operator ends up verifying a key, not an owner. The distinction is not academic. A signature proves control at an instant. Ownership is a claim about intent that persists across time. The rule asks for the second, and the technology can only deliver the first. The revolutionary assumption embedded in the proposal is that a point-in-time proof generalizes into a durable identity property. It does not.

The outbound leg is worse. To release a withdrawal to a wallet, the operator must confirm that the destination wallet belongs to the customer. But the destination wallet sits outside the operator's perimeter. The operator has no authority over it and no way to interrogate it. It can only ask the customer to assert ownership — and an assertion is not a verification.

There is a workable pattern here: enroll destination wallets in advance, run the same challenge-response ceremony before whitelisting, then treat the whitelist as the compliance boundary. Most exchanges already run a version of this. The proposal, in effect, hardens an operational practice into a rule. That is a defensible outcome.

It carries two costs. A pre-enrolled whitelist means a customer cannot send to a new address in the same session — for someone paying a merchant, or repositioning collateral during a volatile hour, the ceremony sits directly on the critical path. And the whitelist becomes the compliance artifact, the thing an examiner reads. Its integrity is now systemically important. A compromised whitelist is not fraud against one customer. It is fraud against the control framework.

The THB 5 million daily ceiling raises questions of arithmetic rather than policy.

Is the cap gross or net? If a customer deposits THB 3 million and withdraws THB 3 million on the same day, has THB 3 million been consumed or THB 6 million? Gross and net differ by a factor of two. Without a defined measurement basis, two licensed operators can implement the same rule and produce different answers for identical activity — and an examiner comparing them will not be able to tell which is correct, only that they disagree.

Is the cap per customer or per account? A customer holding accounts at three licensed operators has either THB 5 million or THB 15 million of daily capacity, depending on the reading. Per-customer enforcement requires a shared registry across operators: a cross-institutional identity layer that does not exist today. Building one is a far larger intervention than the consultation acknowledges. It would also create a centralized map of stablecoin flows bound to national identity documents. That artifact is a more attractive target than any single exchange's database.

There is a pattern here that recurs across financial engineering. DeFi interest rate models — I took Compound's apart in 2020 — are routinely described as market-responsive. They are not. They are parameterized curves selected by governance and adjusted when the curve misbehaves. A measurable proxy gets adopted because it is measurable, and the measurement is then mistaken for the mechanism. A daily cap is easy to specify and easy to audit. Whether it maps to the risk it claims to address is a separate question, and the consultation does not answer it.

The market maker exemption is the most informative line in the paper. Regulators exempt what they cannot afford to restrict. Market makers supply the liquidity that keeps a licensed venue functional. Applying the same-owner test to a firm that continuously moves between its own addresses and venue accounts would break its operations, and with them the depth of the book. The SEC needs the book more than it needs universal coverage.

So the control has a hole cut in it, sized to the volume of the participants whose activity most resembles the behavior the control is meant to catch. That is not a criticism of the SEC. It is an acknowledgment that the rule was never designed to cover the whole perimeter. It was designed to cover the part of the perimeter where coverage is administratively possible.

Which brings the exemptions and the same-owner test into tension. If transfers within certain licensed operator businesses are exempt, and market maker transfers are exempt, and Bank of Thailand-authorized entities are exempt, then the exemption schedule is doing at least as much work as the test. The consultation leaves the interaction between those clauses underspecified. Through a comment period running to September 25, 2026, that underspecification is the thing to watch. It will either be tightened, closing the arbitrage, or left loose — in which case the market will find the seam, because the market always finds the seam.

Here is the counter-intuitive reading. The proposal's problem is not that it is too strict. It is that its strictness binds only where behavior is already observable.

A same-owner test can only be enforced by a licensed intermediary. Unhosted-to-unhosted transfers sit outside the perimeter and outside the enforcement surface. The rule therefore raises the cost of the supervised path and leaves the unsupervised path exactly as it was. Price theory says volume migrates toward the cheaper substitute. AML theory says the volume that migrates is the volume the regulator most wanted to see.

This inverts the usual complaint about over-regulation. The rule does not catch too many legitimate transactions while catching the bad ones; it catches legitimate transactions on a channel it can see, while the channel it cannot see absorbs whatever the cost increase pushes toward it. An AML control that raises the cost of the compliant channel and leaves the non-compliant channel unchanged does not reduce risk. It relocates risk — and it degrades the supervisory dataset in the process.

There is a second-order effect worth naming. Supervised flow data carries intelligence value beyond individual transaction screening. It exposes typologies, counterparty networks, structuring patterns. Every basis point of volume that leaves the licensed perimeter is a basis point of visibility that leaves with it. A rule that shrinks the observable dataset in order to discipline behavior inside it trades detection capability for the appearance of control.

Three things to track between now and the final rule. Whether the daily ceiling is defined gross or net, and per customer or per account — the answers determine whether the limit is one control or three. Whether ownership attestation extends to contract wallets, delegated signing authority, and MPC-secured addresses; if it does, the rule needs an attesting party, and no such party exists on-chain. And whether the exemption schedule becomes a licensable asset, a registry that confers operational advantage on whoever holds a place in it.

If a wallet is proven yours only by producing a signature, and a signature can be delegated, automated, or split three ways, then what has the regulator verified — the owner, or the key?