How to Threat Model Tokenized Treasuries as DeFi Collateral

A practical framework for reasoning about the trust assumptions behind lending against tokenized Treasury and money market fund shares.

The Rise of Tokenized Treasuries as Collateral

Tokenized U.S. Treasury and money market fund products have grown into a significant onchain asset category. RWA.xyz tracks the market across products and issuers. Market size figures should always be tied to a dated snapshot because both the number of products and their total value change over time.

Since mid 2025, tokenized fund shares have increasingly been used in collateral arrangements. BlackRock's BUIDL has been accepted in institutional off exchange collateral programs, while Circle's USYC and Franklin Templeton's Benji funds have also been integrated into institutional collateral arrangements. Other tokenized Treasury products, including Superstate's USTB and funds offered through wrappers, provide further examples of how fund exposure can enter lending and trading systems. These arrangements differ in whether the token is held directly, used through a wrapper, or accepted as collateral by a centralized intermediary. The distinction matters because transfer restrictions, redemption routes and responsibility for execution may differ in each case.

Wrappers such as Securitize's sBUIDL, BUIDL backed USDtb, and funds such as Ondo's OUSG carry exposure through structures that may differ from direct ownership of fund shares. Each layer introduces its own transfer rules, redemption conditions and administrative dependencies.

Research published by Crystal Intelligence in September 2026 examined holder data across 17 large tokenized funds representing approximately $18 billion in Treasuries, private credit and gold. The study found that $3.23 billion of these assets was held in DeFi protocol contracts. Of that amount, approximately $1.12 billion was posted as collateral by third parties, while $2.11 billion was held by protocols as their own reserves. These figures describe the funds in the study, not the entire tokenized asset market.

Once a fund share becomes collateral, the lending protocol's security depends on more than the token contract. It may depend on when the fund calculates and reports NAV, how an oracle handles delayed or rejected updates, whether the token can be transferred to an eligible liquidator, and how quickly the collateral can be redeemed or sold. These dependencies often sit outside the lending protocol's code.

This article presents a threat modeling method for teams considering tokenized Treasury or money market fund shares as collateral. It uses publicly documented fund and redemption mechanisms as examples, while keeping deployment specific assumptions separate from general conclusions. The method follows the full position lifecycle, identifies each trust boundary, and distinguishes what a contract enforces from what the system assumes. It builds on the approach used in our work on threat modeling payment flows on stablecoin native chains.

What a Tokenized Fund Share Secures in a Lending Pool

Mechanically, a loan against a fund token is an ordinary lending position: tokens are supplied, a receipt is minted, an oracle reports a price and a health factor is computed. What differs is what the balance means. It records a claim on a fund share, and whether that claim can be valued, seized or redeemed on the protocol's timeline depends on parties outside the contract. IOSCO, discussing tokenized funds including BUIDL, describes the role of offchain records and transfer agents in determining ownership.

The review therefore has to follow the collateral through its whole lifecycle, not stop at the token contract: price, pledge, liquidate, redeem and govern.

A builder should answer several questions explicitly. Is collateral value the last reported NAV, or what a liquidator could realize today? If the NAV feed stops updating, should the market keep lending, keep liquidating, or freeze? If a liquidator receives fund tokens onchain, when does the transfer agent's register catch up? If the fund corrects a NAV after liquidations have already executed at the old price, who bears the difference?

The threat model should protect five things: a valuation current enough to support lending, collateral that leaves a position only through withdrawal or valid liquidation, eligible holders at every step, a liquidation someone can actually execute, including an exit to cash on the timeline the protocol assumes, and limits on the parties who can change any of these conditions.  These can be written as security properties, which we will refer back to by number.

P1: a price the protocol cannot establish as sufficiently current and reliable does not create borrowing capacity.

P2: collateral leaves a position only through withdrawal or valid liquidation, except for explicitly modeled administrative or recovery paths.

P3: every address that ends up holding the asset is eligible to hold it.

P4: every liquidatable position has someone able and willing to liquidate it in time.

P5: privileged actions cannot silently invalidate the protocol's assumptions about collateral ownership, value or accounting.

Each property can fail without a single contract call reverting.

Price, Exit and Authority: Three Paths Through the Position

Several parties that can affect a position never touch user funds directly. An issuer can pause a token or remove an address from its allowlist. A fund's redemption calendar can push an exit back by days. An administrator can publish, delay or correct a NAV, and an oracle network can filter it. The transfer agent's register may be the legal record of ownership, even when token transfers occur onchain.

The first path is the price path. It may include the fund administrator computing NAV, the issuer reporting it, an oracle network deciding whether to publish it, an adapter transforming the answer, and the lending protocol using it. Each stage can introduce a different failure mode. A contract may store a price while the validation rules that determine whether it is updated operate offchain. The threat model must establish where each check runs, who controls it and what happens when a check rejects an update.

The second path is the exit path, from a health factor breach to cash in a liquidator's hands. It includes liquidator eligibility, access to the debt asset, the ability to receive the fund token, a conversion route, fund settlement and the banking rail. Some steps are enforced by contracts. Others depend on offchain counterparties, liquidity and operating hours.

The third path is the authority graph. Its nodes are the contracts and registers that record a position: the fund token and its compliance services, the transfer agent's book, the lending pool and its receipt token, the oracle, and any wrapper. Its edges are powers such as pause, seize, burn, force transfer, upgrade, set a price or correct a record. Each power has a holder, a threshold, a delay and a scope.

For each path, list every handoff between parties: administrator to oracle, adapter to pool, pool to compliance check, liquidator to redemption facility, and onchain transfer to transfer agent register. At each handoff, separate what a contract enforces from what agreements, operators or key custody are assumed to provide. The latter belongs in the threat model as a named dependency with a defined failure mode.

Five Assumptions a Lending Market Makes About Fund Tokens

The following assumptions describe risks that can arise when a lending market accepts tokenized fund shares as collateral. The examples are threat scenarios, not claims that a particular incident has occurred. Each assumption ends with the properties it puts at risk.

Assumption 1: Collateral Is Priced on the Fund's NAV Schedule

Many fund tokens do not trade in a deep secondary market, so their onchain price may depend heavily on the NAV reported by the fund. USYC, for example, documents its daily pricing process. A daily NAV is not inherently unsafe, but its update schedule determines how quickly the lending market can reflect changes in the underlying asset.

Start with the oracle interface. Determine whether the lending protocol checks the answer's timestamp, handles invalid or nonpositive answers safely, and defines what happens when a feed stops updating. A fallback oracle is not automatically safer. The threat model should establish which source is authoritative, how a fallback is activated, and whether either source can return a value that is technically valid but economically stale.

Next, examine controls outside the lending contract. An oracle network may reject a reported NAV that falls outside configured bounds. This can prevent an erroneous update from reaching the lending market, but it can also delay a genuine loss from being reflected in the price. If rejected updates leave the last accepted price in use, the protocol may continue to calculate borrowing capacity from a value that no longer represents the collateral's current economic condition.

The threat model should establish who monitors rejected updates, how the rejection is surfaced, and how quickly borrowing can be restricted while the price is investigated. A stale price can be conservative when the asset steadily accrues value, because it may understate collateral value. The more dangerous case is a stale price after a genuine loss.

A fund targeting a stable $1 NAV illustrates a different limitation. BUIDL's NAV is described as generally expected to be $1.00. A reported NAV may remain unchanged even when redemption capacity, settlement time or the amount of cash available through a particular exit route changes. A stable reported price is therefore not sufficient evidence that the collateral can be realized at that price on the timeline assumed by a lending market.

NAV corrections create a related problem. A fund's correction process may not reverse transfers or liquidations that have already taken place onchain. The protocol should establish how it handles a corrected price, whether positions can be liquidated at the wrong price before the correction arrives, and who bears any resulting loss.

Puts at risk: P1. The protocol must not extend borrowing capacity based on a price it cannot establish is sufficiently current and reliable. Timestamp checks, invalid answer handling, oracle safeguards and emergency controls may operate at different layers and must be evaluated separately.

Assumption 2: To the Issuer, the Pool Is a Single Shareholder

When fund tokens are supplied to a lending pool, the issuer may see the pool address as the holder even though the position represents collateral for many borrowers. The protocol's internal accounting can distinguish borrowers, but the issuer's compliance system may apply restrictions to the pool's aggregate token holding.

This creates a mismatch between the lending protocol's internal positions and the issuer's view of ownership. Lockups, transfer restrictions and eligibility checks can prevent a transfer even when the lending protocol considers the position liquidatable. The exact checks and their consequences must be established for each asset by inspecting the relevant token contracts and governing documentation.

Distributions create another issue. If a fund pays yield by issuing additional tokens, those tokens may arrive at the pool's address rather than being credited separately to each borrower. The protocol must define how such distributions are accounted for and how they affect collateral value and borrower balances.

A pooled position also creates a compliance boundary. The protocol may need to respond to a sanctioned borrower or an issuer's transfer restrictions without making every other borrower ineligible to withdraw or liquidate. Any privileged mechanism that moves a borrower's position or redirects collateral should be examined for recipient eligibility, authorization, accounting effects and auditability. If the mechanism does not enforce the relevant eligibility checks in code, the threat model should identify the operational process relied on instead.

Wrappers introduce a separate boundary. A wrapper may hold the fund token while issuing another token that is transferable under different rules. The wrapper's redemption rights, backing, transfer restrictions and administrative powers must therefore be evaluated separately from those of the underlying fund. The wrapper should not be assumed to preserve every protection of the underlying token automatically.

Puts at risk: P3 if a transfer path allows an ineligible address to receive the asset, and P2 if issuer level restrictions or privileged actions can alter a pooled position outside the protocol's expected withdrawal or liquidation paths.

Assumption 3: Only Eligible Liquidators Can Liquidate

In a permissioned collateral market, the set of eligible liquidators may be smaller than the set of accounts that can interact with the lending contract. If liquidation requires transferring the underlying fund token, the liquidator must satisfy the issuer's eligibility and transfer rules and have access to the debt asset needed to complete the transaction.

Those conditions may be separate from the protocol's health factor calculation. The contract can identify an unhealthy position, but it cannot guarantee that an eligible liquidator has sufficient capital, can receive the collateral or can execute the transaction economically.

The threat model should also establish how the protocol accounts for a position that cannot be liquidated promptly. Interest may continue to accrue while the position remains unhealthy, and a loss may not be recognized immediately under the protocol's accounting rules. The exact accounting treatment depends on the implementation and should be verified against the relevant contracts.

Liquidator capacity should be evaluated against the amount of debt that could become liquidatable under a plausible adverse NAV change, rather than against an arbitrary percentage of total borrowing. Where many positions use the same daily NAV, the analysis should consider how much debt could become liquidatable at the same time.

The operational review should identify each liquidator, its eligibility status, committed capacity in the actual debt asset, expected response time, settlement dependencies and fallback arrangements. Informal indications of capacity are not equivalent to a binding commitment or a service level agreement.

Puts at risk: P4. Liquidation depends on eligible counterparties and their actual capacity, not just on the lending protocol's ability to calculate a health factor.

Assumption 4: "Instant" Redemption Is a Capped Buffer With an Operator

A liquidator's willingness to act depends partly on how it expects to exit the collateral. Tokenized funds may offer more than one route, but those routes can have different eligibility requirements, cutoffs, minimum amounts, capacity limits and settlement times.

BUIDL illustrates the difference between a fund's USD redemption process and a separate route for exchanging fund shares for USDC. Under the terms described in Securitize Markets' public application, USD redemptions depend on the transfer agent recording the relevant shares by the stated business day cutoff. The application describes a $250,000 minimum for partial redemptions. Because the application dates from 2024 and includes later updates, current terms should be verified against the fund's governing documents before relying on exact timing or minimum amounts.

The separate Circle BUIDL off ramp provides another way to exchange BUIDL for USDC. The public application describes this route as available at Circle's discretion and not guaranteed. It is a separate route, not a guarantee that the fund itself can always redeem tokens instantly.

Other products also distinguish between faster exit facilities and their underlying redemption process. USYC, OUSG and USTB have product specific redemption arrangements. Their eligibility requirements, available capacity, settlement rules and contract behavior should be reviewed separately. A capacity limit documented for one product must not be treated as a general feature of tokenized Treasury funds.

The threat model should assume that a fast route may be unavailable, fully used or unable to process the amount required during a stress event. It should then establish whether the official redemption route is still available and how long it takes to deliver cash. Business day cutoffs, transfer agent processing and banking holidays can all affect the timeline.

Puts at risk: P4 indirectly. A liquidator may be unable or unwilling to act if it cannot establish a sufficiently reliable exit for the collateral.

Assumption 5: Authority Is Layered Across Issuer, Transfer Agent, Protocol and Wrapper

Every layer around a position may hold powers that can change it without asking the borrower. These powers can be necessary for sanctions compliance, court orders, lost keys, corrections or emergency response. The security question is whether the protocol has accounted for them.

Depending on the token and its implementation, issuer level powers may include pausing transfers, restricting addresses, seizing or burning tokens, forcing transfers or upgrading a contract. The transfer agent may also have authority to correct its ownership register or reject transactions that do not meet the governing rules. The legal and operational consequences depend on the fund's documentation and token implementation.

The protocol layer adds its own administrative powers. Depending on the design, these may include emergency freezes, reserve configuration, asset listing, authorized transfers and oracle parameter changes. Governance delays and multisignature approval requirements can constrain these powers, but their effectiveness depends on the permissions actually granted and the independence of the controls.

A complete threat model should map each privileged role to its permitted actions, approval threshold, execution delay and scope. It should also identify where multiple roles share an administrator, signer or underlying authority. Controls that appear independent at the contract level may not provide independent protection if the same authority can exercise both.

For each privileged action, document the intended security property, the conditions under which the action is permitted, and the accounting and compliance consequences if it is exercised against a pooled position. Test emergency freezes, authorized transfers, token pauses, seizures and upgrades against the full lending lifecycle.

Puts at risk: P2 when privileged actions change collateral outside the expected withdrawal or liquidation paths, and P5 when administrative powers can invalidate the protocol's assumptions about collateral ownership, value or accounting.

When the Assumptions Fail Together

The assumptions can interact, but the consequences depend on the size of the loss, the remaining collateral buffer and the availability of an executable exit. The following is a composite threat scenario, not a description of a single deployed market. It combines the general risks of NAV based oracle safeguards with redemption constraints documented for tokenized fund products.

Suppose a fund reports a genuine loss that is not reflected in the price used by a lending market. This could happen if an oracle network rejects the update as outside configured bounds, or if the reported NAV does not yet reflect a change in the amount that can be realized through redemption. If the lending market continues to use the last accepted price, borrowers may retain borrowing capacity based on a valuation that no longer reflects the risk the protocol needs to manage. This violates P1 even though the oracle keeps returning a formally valid answer and no contract call reverts.

The size of the loss matters. Take illustrative numbers: an oracle that rejects NAV updates moving more than 0.05%, and positions at 88% LTV against a 90% liquidation threshold. Those positions become liquidatable only after the collateral loses about 2.2% of its value, so a small loss just above the oracle’s bound does not push them over. A loss larger than that does, and positions with similar LTVs cross their thresholds together. If the loss is genuine and the oracle keeps rejecting it, the protocol keeps pricing the collateral at the old NAV the whole time. How much debt becomes liquidatable at once depends on how health factors are spread across the market and on the size of the NAV change.

Even when a position becomes liquidatable, execution is not guaranteed. If only eligible liquidators can receive the underlying fund token, they must meet the issuer's transfer requirements and have access to the required debt asset. They must also be able to exit the collateral through a redemption or conversion route that is available and sufficiently funded. A delayed or unavailable exit can make liquidation uneconomic or prevent it from completing in time, putting P4 at risk.

A separate scenario concerns a fund targeting a stable $1 NAV. Its reported price may remain at $1.00 while a redemption restriction, liquidity fee or settlement delay changes the amount of cash that can be realized, or the time required to realize it. In that case, P1 is violated if the protocol continues to extend credit against a valuation that no longer represents the value it can safely rely on. If no eligible liquidator can execute the position and realize the collateral in time, P4 may also be violated.

These scenarios illustrate why the threat model must assess price validity, liquidation capacity and redemption availability together. The relevant controls are the ability to prevent additional borrowing against an unreliable price, the speed of emergency intervention, and the amount of eligible liquidation capacity available under adverse conditions. The outcome cannot be inferred from the oracle tolerance or LTV alone.

Tokenized Treasury Collateral Checklist for Builders and Auditors

The checklist follows the distinction between what a contract enforces and what the protocol depends on. The first group can be established in code and in an audit. The second requires operational evidence, named owners and, where possible, written commitments.

Verify in Code

  • Write P1 to P5 as invariant tests or formal specifications for the lending contracts, and record which properties the deployed design enforces and which it leaves to operations.
  • Test every feed failure mode through the full adapter path: holding the last value, returning an end of range price, extrapolating, reverting after an expiry, and returning zero. Include the effect of a revert on positions with other collateral. Check decimals and scaling on the final price.
  • Measure staleness from the NAV's own date, sourced outside the adapter where the adapter exposes no timestamp, and verify that a price failing that check cannot create borrowing capacity.
  • Record where each protection executes. Bounds and circuit breakers run by an oracle network or monitoring desk are not contract properties.
  • Verify that authorized and forced transfers check the recipient against the underlying allowlist, and that no path leaves collateral or debt with an ineligible address.
  • Test liquidation against every relevant rule in the token's compliance service, including removal of a liquidator, borrower or pool from the allowlist, lockups and an issuer pause. Include interest accrued while liquidation is unavailable.
  • Test the pool's accounting for distributions that land at its address and for a seizure, burn or forced transfer against its holding.
  • Build an inventory of pause, freeze, seize, burn, upgrade, force transfer and oracle setter powers from contract code and onchain role events. Record holders, thresholds, delays and scope. Cross reference holders across layers and flag privileged grants that remain active.

Confirm as Dependencies

  • Name each asset's eligible liquidators and their committed capacity in the actual debt asset. Size that capacity for a defined adverse NAV scenario, including the amount of debt that could become liquidatable together, and set supply caps within the capacity that can reasonably be relied on.
  • For each redemption route, record the operator, cap, funding source, cutoff, minimum and settlement cycle. Confirm that expected liquidation amounts can meet the minimums, and keep separate calendars for when NAV is struck, when redemptions are accepted and when cash can move.
  • Agree with the issuer and transfer agent how the pool is registered, how quickly an onchain liquidation is reflected in the register, whether a transfer can be unwound, whether pauses are token wide or address specific, and how NAV corrections, liquidity fees and redemption delays will be communicated.
  • Name who monitors fund notices, how quickly borrowing can be restricted or a reserve frozen, and who is responsible for escalation. Establish how NAV corrections and bad debt are handled.

Four Questions Every Collateral Listing Should Answer

Tokenized Treasuries bring useful building blocks onchain, but the lender relies on the collateral guarantee that the protocol builds around them. For every collateral asset, the threat model should answer four questions.

What is the collateral worth right now, and how old is that number? Who can take it, whether as a liquidator, an issuer or a transfer agent? How quickly does it become cash, and through whose facility? Who can stop that process, and under what conditions?

If you are building a lending market, vault or wrapper around tokenized Treasuries or other tokenized funds, Certora can help with smart contract audits, formal verification of onchain components, and threat modeling of the assumptions connecting fund structure, oracles, issuer controls, liquidators and redemption paths. Reach out to discuss an architecture review, audit engagement or longer term security support. Follow Certora on X and LinkedIn for more threat models and security research.

FAQ

Can tokenized Treasuries be liquidated onchain?

Yes, where a lending protocol supports the token and its transfer rules allow an eligible liquidator to receive it. Whether the liquidator can then turn the token into cash depends on the fund's redemption or conversion routes, including their eligibility requirements, capacity and settlement times.

How are tokenized Treasuries priced in DeFi lending?

A lending protocol may use a fund's reported NAV through an oracle feed. The threat model should establish how frequently that value is updated, whether timestamps and invalid answers are checked, how rejected updates are handled, and whether borrowing can continue against a stale value.

What happens if the issuer pauses its token?

The effect depends on the token's implementation and the operation being attempted. A pause may prevent transfers needed for supply, withdrawal or liquidation. The protocol should test each relevant operation and establish how it manages positions while transfers are unavailable.

Who bears the loss if a position goes underwater?

The answer depends on the lending protocol's liquidation and deficit accounting rules. An unhealthy position does not necessarily mean that a loss has already been recognized in the protocol's accounting. Review the implementation to establish when a deficit is recorded, which assets or participants absorb it, and whether a backstop or recovery mechanism exists.

Which tokenized Treasuries are accepted as collateral?

Acceptance depends on the specific lending protocol or collateral program and can change over time. Check the protocol's current supported asset list and the issuer's eligibility requirements before assuming that a token can be supplied or received by a liquidator. Institutional off exchange collateral arrangements should be distinguished from onchain lending markets.

Get every blog post delivered

Certora Logo
logologo
Terms of UsePrivacy Policy