Custody Key Threats & Controls
Who can make a custody system sign, and which control combinations stop unauthorized signing today — blind vs clear signing, exfiltration paths, adversary models, case studies, and hybrid MPC + HSM architectures.
Why this matters: Keeping private keys inside an HSM, MPC cluster, or enclave does not prevent unauthorized transfers if an attacker can deceive the signing screen, hijack an authorized requester, or abuse pre-signed authorizations. Decisions depend on putting transaction parsing, policy checks, and quorum approvals inside a protected boundary with stated authority assumptions.
Start here: Open Blind vs Clear Signing, sign a sample transfer hash with CKM_ECDSA, then swap the hidden payload for a delegatecall upgrade to see why parsing without a policy check refuses nothing.
Educational scope & service-vs-asset boundary
This module uses fictional wallet scenarios and browser-simulated cryptographic runs to teach custody decision-making; completing these lessons or workshops does not verify or certify any deployed custody platform. In institutional custody, distinguish business services (such as a custody signing service or a long-lived records archiving and retrieval service) from protected assets (private keys, threshold shares, digital assets, and retained audit records themselves).
A custody design fails when teams treat where the private key sits as the entire security model. Holding a key inside an HSM or MPC cluster prevents direct raw-key export, but an attacker who controls signing authority — the ability to construct a payload, deceive the approval screen, reuse a pre-signed message, or trigger an API credential — can move funds without ever extracting the private key.
Foundational curve mechanics (secp256k1 vs Ed25519), deterministic ECDSA nonce derivation under RFC 6979, HD wallets, and the five-layer custody stack are taught in Digital Assets (LM-045) and are linked rather than repeated here. This module begins where key placement ends: who can make a custody system sign, and what stops unauthorized authority.
ECDSA requires a fresh, random, uniformly chosen per-signature nonce k; RFC 6979 defines deterministic nonce generation so verifiers need no changes while avoiding RNG reuse failures.
Source: IETF RFC 6979 (Informational, August 2013), §1 Introduction (pp. 2–3)
In Safe.sol documentation, checkNSignatures verifies caller-supplied owner signatures over a transaction hash and warns callers to ensure executor is legitimate because an incorrect executor can reduce the required signature count by one.
Source: Safe{Wallet} Smart Account Docs, checkNSignatures -> Parameters -> executor
In Safe.sol contract source, checkSignatures enforces that the stored threshold is nonzero (reverting with GS001 otherwise) before passing that stored _threshold to checkNSignatures.
Source: Safe Smart Account Safe.sol Source, contracts/Safe.sol, checkSignatures
EIP-7702 introduces an EIP-2718 transaction type that lets an Externally Owned Account (EOA) attach signed authorization tuples and write a persistent delegation indicator (0xef0100 || address) into its account code.
Source: Ethereum EIP-7702 (Pectra), Abstract & Specification
Rule G1 (Authority Assumptions)
Every claim in this module that a control “holds” explicitly states which signer, guardian, recovery, and contract-upgrade authorities an attacker must not hold for that guarantee to remain intact. Note how checkNSignatures depends on caller-supplied threshold and executor arguments unless gated by checkSignatures reading storage, and how EIP-7702 lets an EOA delegate execution persistently without exposing its private key.
Check off all sections and mark this reading done.
Related modules
- Custody Key Lifecycle OperationsSame track · Industries · Same migration phase · Shares ECDSA, ML-DSA, FIPS 140-3
- Custody PQC MigrationSame track · Industries · Same migration phase · Shares ML-DSA, Falcon, ECDSA
- Energy, Utilities & Grid PQCSame track · Industries · Same migration phase · Shares ML-DSA, ECDSA, FIPS 140-3
- Digital AssetsSame track · Industries · Same migration phase · Shares Falcon, ML-DSA, FIPS 140-3
Check your understanding
55 questions on Custody Key Threats & Controls, each with its answer and the reason.
Take the quizNext step
Produce the artifact: Deployment PlaybookThis module belongs to phase 5 (Pilots & Migration); Deployment Playbook produces a deliverable of that phase in the Command Center.
Production Tool Capabilities & Verification ChecklistCustody Security Architects, Operations Leads & Risk Governance Teams
Capabilities to require in your production tools:
- In-boundary transaction parsing, allowlist checks, and m-of-n quorum enforcement executed inside the HSM, enclave co-signer, or attested approver boundary rather than a host application.
- Independent trusted-display clear signing (e.g., ERC-7730 descriptors paired with bounded ERC-8176 attestations) and bounded approval expiry/nonce binding preventing replay or silent delegatecall upgrades.
- Explicit Rule G1 authority separation across signing shares, orchestrators, recovery quorums, and contract-upgrade admins (avoiding single-operator or host-RAM key unwrap exposure).
How to evaluate in your environment
Exercise staging custody policies against negative test cases: tampered amounts, expired approvals, replayed nonces, unallowlisted delegatecalls, and single-party threshold refusal.
Production evidence to collect
Signed custody policy ruleset, FIPS 140-3 CMVP module certificate & Excluded Components boundary review, and staging negative-test rejection logs.
Completing this Hub module and its browser-based sample exercises teaches the technical and governance decision criteria for your role; it does not test, validate, or certify your deployed production systems.
Record evidence for your environment
Identify the tested build and configuration. Use reference IDs for observations. For a proposed exclusion, record the system boundary or rule and why it excludes this build.
Evidence missing. 0 of 3 requirements have a reference or scope rationale.
Learning module content can be inaccurate. Please double-check its information. Report inaccuracies in PQC Today GitHub Discussions.