For a yield-bearing tokenized treasury product where confidential investor positions and a complete transfer-agent register are the load-bearing properties, ZK Shielded Commitments is the default: positions and redemptions are private, and aggregates are published at the fund's cadence rather than per transaction. FHE suits funds whose existing integrators and contract holders must keep operating on balances, since its account model keeps addresses and hides amounts only; the trade-off is reliance on the FHE network for both decryption and throughput. TEE is a viable PoC starting point and a near-term production option for funds whose custodians already accept hardware-rooted trust.

The dominant engineering questions are yield distribution into hidden positions and the size of the anonymity set. ZK with a running total_shares commitment keeps per-transaction proving constant and pushes full-audit cost off the critical path; circuit complexity for gate enforcement and yield attribution is the ceiling. FHE simplifies the programming model but inherits shared-throughput limits and per-ciphertext revocation gaps. TEE eliminates both proving cost and revocation issues but introduces a hardware trust chain and side-channel surface that auditors must accept. Threshold custody administration (key rotation, custodian onboarding, threshold parameter selection) is non-trivial across all three.

This is a perspective for legal review by the deploying fund operator, not legal advice. The three options expose distinct evidence patterns: ZK Shielded Commitments via per-position view keys plus ZK public outputs proving investor-side gate compliance (eligibility, jurisdiction caps; portfolio rules are measured on the fund's assets) without revealing positions, with EAS-logged disclosures as the trail; FHE via per-balance ACL granularity with no per-ciphertext revocation (revocation depends on subsequent balance updates triggering re-grants, or the disclosure model has to rely on append-only audit logs); TEE via enclave-mediated disclosure that depends on enclave-code reproducibility, multi-party signing, and the vendor governance model. Whether any of these patterns satisfies SEC Rule 2a-7 or ESMA MMFR auditor expectations is a question for jurisdictional review. Custody of disclosure keys (the transfer agent's register key, regulator viewing keys; independence and jurisdictional diversity where threshold-held) is a structural-risk parameter that legal review would weigh in any of the three.

Requirements

  • Daily or intraday NAV computation with verifiable correctness (total shares outstanding stays public)
  • Aggregate figures are published at the fund's cadence, not per transaction, so totals don't reveal individual flows
  • SEC Rule 2a-7 (US) and ESMA MMFR (EU) compliance: liquidity fees, concentration limits, and MMFR liquidity gates. Private funds instead follow offering requirements such as investor eligibility and investor-count limits.
  • Atomic subscription and redemption settlement (no partial fills: shares and cash move together or not at all). Where the cash leg settles off-chain, the transfer agent reconciles.
  • Yield attribution provably correct per investor without revealing positions
  • The transfer agent can rebuild the full register of record for any point in time (eligibility, holder counts, sanctions screening, tax) without holder cooperation
  • Per-holder choice: confidential by default, with opt-out for holders that must stay public
  • Eligibility (KYC, accreditation, jurisdiction) is proven without a public wallet-to-investor mapping

Constraints

  • Custody of disclosure keys must be administratively feasible for the transfer agent and regulators
  • Periodic full-audit checkpoints must run off the critical path of subscription/redemption
  • Fund-circuit hash registered immutably at deployment; circuit upgrades imply migration
  • No personal data on-chain, even encrypted, because data-protection rules such as GDPR conflict with an immutable ledger: the register feed carries pseudonymous investor IDs, and personal data stays off-chain with the transfer agent
  • Compliance gates (Rule 2a-7 liquidity ratio, concentration limits) must be enforceable without revealing individual positions. These portfolio rules apply to the fund's assets, so investor privacy doesn't affect them; investor-side rules such as liquidity fees depend only on aggregate flows.

Architectural options

ZK Shielded Commitments

documented

Share positions are shielded UTXO commitments; a running total_shares commitment is updated per transaction; the running total is opened at the fund's publication cadence by the transfer agent, and anyone can check the opening against the on-chain commitment.

How it works

Each position is a commitment to (attestation hash, share count, and entry NAV for floating-NAV funds). Subscription mints a position commitment and increments a running Pedersen commitment to total_shares; redemption nullifies the position and decrements the running total. ZK circuits enforce conservation, gate logic (investor-side: eligibility, jurisdiction caps, investor counters; the Rule 2a-7 liquidity ratio and concentration limits are portfolio rules, enforced outside the pool), and yield-attribution constraints. At the fund's publication cadence, the transfer agent, which receives each transaction's opening through the register feed, opens total_shares and publishes the total, which anyone can check against the on-chain commitment; the fund publishes NAV per share; a periodic full-audit checkpoint verifies the running total against all active positions, off the redemption critical path.

Trust assumptions
  • L1 / L2 consensus and verifier contract correctness
  • Transfer agent for opening and publishing the running total (it already holds the full register); the register key can be threshold-held with independent co-holders
  • Oracle integrity for per-share price (floating-NAV funds; single oracle or quorum)
Threat model
  • Adversary observes L1 / L2; cannot break ZK soundness
  • Register-key compromise reveals positions to the attacker; the published total reveals only aggregates at the fund's cadence
  • Oracle compromise distorts NAV (floating-NAV funds); mitigated by quorum
  • Periodic full-audit catches running-total drift from circuit bugs
Works best when
  • Investor positions must be hidden from the public while the transfer agent keeps the full register
  • Daily or intraday NAV cadence matches the proving budget
  • Disclosure-key custody (register key, regulator viewing keys) is administratively feasible
Avoid when
  • Yield logic is complex enough that circuit complexity exceeds practical bounds
  • The confidential set is small and concentrated, so unlinkability gains are limited
Implementation notes

The PoC uses the shielded pool in 2/SHIELDED-POOL with attestation-gated entry (3/ATTESTED-POOL), as implemented in EthSystems' shielded-pool-compliance: KYC-gated entry, attestation expiry, circuit-level compliance checks, and an encrypted audit channel. The MMF variant adds investor positions, yield distribution, investor counters, ZK compliance outputs, per-position regulator view keys, and yield as new shares (a periodic mint or a multiplier; pro-rata attribution applies only to floating-NAV funds), with fund-level portfolio rules measured outside the pool.

FHE Encrypted Balances

documented

Balances are FHE ciphertexts handled by an FHE network over the host chain (for example, Zama on Ethereum mainnet); NAV is computed homomorphically; threshold key holders decrypt for publication.

How it works

Subscriptions encrypt the share count under the FHE network's keys; balances are stored as ciphertexts with ACL-based read access. NAV is computed under encryption (sum of ciphertexts × per-share price); a t-of-n threshold network decrypts the result for posting on chain. Yield attribution runs as homomorphic arithmetic; gate logic uses encrypted comparisons.

Trust assumptions
  • t-of-n threshold network for FHE decryption keys
  • FHE library implementation correctness
  • ACL grants honored by all participants
Threat model
  • Threshold compromise reveals all balances
  • No revocation per ciphertext; ACL revocation requires re-encryption on balance update
  • Shared throughput across all FHE applications on the network is a bottleneck
Works best when
  • Existing integrators and contract holders must keep operating on balances (account model), and amount-only confidentiality is acceptable
  • Yield logic is complex (path-dependent strategies) and benefits from homomorphic arithmetic
  • Per-balance ACL granularity matches the disclosure model
  • Threshold-network trust is acceptable to all custodians
Avoid when
  • Operator independence requires that no single network be load-bearing
  • Throughput pressure across all FHE applications is unacceptable

TEE Enclave

documented

Positions sealed inside a TEE enclave; NAV computed in the clear internally; remote-attested results posted on chain.

How it works

Subscriptions are encrypted to the enclave's attested public key; the enclave maintains positions in cleartext memory. NAV computation, gate enforcement, and yield attribution all run inside the enclave; the enclave signs results under its attested key. Selective disclosure runs through enclave-mediated views: regulators submit signed queries, the enclave returns scoped responses.

Trust assumptions
  • TEE vendor and remote-attestation chain (Intel TDX, AMD SEV, AWS Nitro, Azure CC)
  • Cloud co-tenant isolation
  • Enclave software supply chain (reproducible build, multi-party signing)
Threat model
  • Side-channel and microarchitectural attacks against the enclave class
  • Vendor compromise or compelled disclosure of attestation keys
  • Enclave software vulnerabilities expose all fund state to a single-party adversary
Works best when
  • Near-term deployment is needed and ZK / FHE proving costs are prohibitive
  • Hardware trust is already accepted by custodians
  • Enclave reproducibility and signing pipeline are operationalized
Avoid when
  • Threat model includes nation-state-class side-channel adversaries
  • Operator independence requires that no single hardware vendor be load-bearing

Side-by-side

Axis ZK Shielded Commitments FHE Encrypted Balances TEE Enclave
Maturity documented documented documented
Context i2i i2i i2i
Trust model Math + transfer agent for publishing aggregates Threshold (t-of-n) decryption Hardware vendor + supply chain
Privacy scope Amounts + addresses (bounded by anonymity set) Amounts only; addresses public Amounts + addresses (inside enclave)
Performance Constant per tx; periodic full-audit scales with positions Heaviest compute; shared throughput Cheapest; near-instant
Operator req. None beyond the transfer agent (relayer optional) Yes, beyond the transfer agent (threshold network) Yes, beyond the transfer agent (enclave host)
Interop with integrators and contract holders Low: replaces the token's execution layer, and contracts can't hold note secrets Medium: address-based, and contracts can hold and move encrypted balances Medium: through the enclave's attested interface
Cost class Medium-high Medium Low
Yield distribution cost Per position per period (mint), or one update (multiplier) Per balance per period (homomorphic add), or multiplier Low: computed inside the enclave
Regulatory fit Strong (per-position view keys, EAS-logged) Strong (per-balance ACL, no revocation) Conditional (vendor attestation)
Failure modes Register-key compromise; circuit bugs; oracle compromise (floating-NAV funds) Threshold compromise; no revocation; throughput Side-channel; vendor compromise; enclave outage

Decision factors

  • If existing integrators and contract holders must keep operating on balances, or per-balance ACL granularity is required, and amount-only confidentiality is acceptable, choose FHE Encrypted Balances.
  • If near-term deployment is required and custodians already accept hardware-rooted trust, choose TEE Enclave as a PoC or transitional path.
  • If the confidential set stays small and concentrated (single fund, few large holders), none of the three approaches delivers meaningful unlinkability; widen the set (shared pool, cash leg) or re-scope to amount confidentiality.

Hybrid composition

Hold confidential positions in ZK Shielded Commitments and keep opted-out holders on the public token, with conservation spanning both. Use TEE to compute per-period yield distribution across hidden positions when it exceeds practical proving budgets, with the TEE output committed back into the ZK state at the next checkpoint. FHE Encrypted Balances serve integrator and contract-holder positions that must stay operable where amount-only confidentiality is acceptable. Compliance gating runs uniformly through the regulator-disclosure-keys pattern across all rails.

Open questions
  1. Yield-cohort attribution (floating-NAV funds). Balancing fungible-share privacy against the complexity of attributing yield across different entry-NAV cohorts; pro-rata is the working assumption but other models may better serve specific products. Stable-NAV funds avoid cohorts by paying yield as new shares.
  2. Peer-to-peer share trading. Shielded MMF shares traded peer-to-peer with privately enforced NAV-based pricing; integration with shielded DvP is unresolved.
  3. Mixed-currency underlying. Hiding currency exposure when the underlying basket spans multiple currencies; oracle and ACL design are unresolved.
  4. MMF-as-cash-equivalent. Whether shielded MMF shares can serve as the cash leg in DvP for other instruments.
  5. Anonymity-set source. Single fund, pool shared across funds, or shared with the cash leg, and how per-fund rules are enforced in a shared pool.
  6. Account vs note. Amount-only confidentiality with interop, vs unlinkability with a replaced execution layer.
  7. Participation. Privacy by default with opt-out; what each public ↔ confidential crossing leaks; a minimum confidential-set size.
  8. Transfer-agent visibility. The minimum the register feed must carry (ownership vs flows), and constraining the ciphertext in-circuit.
  9. Yield into hidden positions. Periodic mint vs multiplier, and whether a multiplier counts as reinvestment for reporting and tax.
  10. Custody. Multisigs, smart accounts and contract holders; client-side vs delegated proving.
  11. Migration from deployed tokens. In-place upgrade vs reissuance, and interop cost per integrator. Several deployed tokenized MMFs run Securitize's DS Protocol, with two generations in production (omnibus removed in the newer one); in-place upgrades must fit each generation's storage layout. Locking the deployed token and mirroring positions on a confidential ledger (origin-locked pattern) makes the lock contract a nominee holder: investor counters go wrong, or the reporting that fixes them is public.
  12. Liveness under stress. The holder's path to cash when an operator or cash venue is unavailable.
  13. Collateral (future extension). Locking a hidden position in favour of a lender, the lender's claim on default without the borrower's secret, a confidential cash leg, and keeping the register of record accurate while shares sit in a lending pool.

Referenced by

Last reviewed