Custody Key Lifecycle Operations
How digital-asset custody keys are created, guarded, watched, recovered, and retired in daily operation — DKG & refresh, rotation vs sweeps, recovery & inheritance, ceremonies, monitoring, access separation, backup & DR, and safe decommissioning.
Why this matters: Custody failures happen across the entire key lifecycle, not only during signing: unrevoked token allowances or EIP-7702 delegations outlive withdrawal pauses, untested recovery paths become attack paths, rushed restores re-enable attacker footholds, and premature key destruction permanently burns incoming funds.
Start here: Open W9 (Revoke or Rotate Matrix) and classify twelve custody items — including an EOA hot wallet, an MPC share, a departed approver device, a token allowance, and an EIP-7702 delegation — to see why sweeping, resharing, and revoking solve different problems.
Educational scope & Rule G1 authority assumptions
This module teaches how custody keys are created, guarded, watched, recovered, and retired in daily operation using fictional scenarios and sample workflows. Completing workshops or checklists here does not verify or certify a production custody deployment — apply these checks inside your own production HSM, MPC, IAM, and monitoring tools.
Rule G1 (authority assumptions): every lifecycle control below states which signer, guardian, recovery, backup, and upgrade authorities an attacker must not hold for the control to hold. Prerequisite: Module A1 (Custody Key Threats & Controls).
5a. Threshold Key Lifecycle (DKG, Refresh, Membership & Availability)
Threshold custody splits signing authority across n parties so any t can sign. How shares are born, refreshed, backed up, and reassigned determines whether a single point of compromise ever exists:
- Distributed Key Generation (DKG) vs dealer import: DKG generates shares jointly so the full private key never materializes on one machine; dealer import creates or imports a full key in one place and then splits it, leaving a single-point compromise window.
- Proactive share refresh & membership change: resharing replaces every party’s share without changing the public key, defeating a mobile adversary who compromises fewer than t shares per epoch and enabling participant additions or removals.
- Backup rebuild risk: backing up shares carelessly — or restoring them onto one workstation — recreates the single point of failure that threshold signing was meant to remove.
- Identifiable abort & availability: a malicious or offline party can stall signing; protocols with identifiable abort attribute the fault so operators can exclude the misbehaving node without lowering threshold security.
NISTIR 8214 surveys threshold cryptographic schemes and notes that distributing cryptographic operations across multiple devices or parties introduces new protocol complexity and can increase the attack surface unless implementation, communication, and validation challenges are addressed.
Source: NISTIR 8214 (March 2019), §1 Introduction & §4 Challenges in Threshold Cryptography
Herzberg et al. (CRYPTO 1995) introduced proactive secret sharing (PSS) so share holders periodically refresh shares of the same secret without changing the secret itself, defeating a mobile adversary who compromises fewer than t shares per epoch.
Source: Herzberg et al., CRYPTO 1995 (LNCS 963), §1 Introduction & §3 Share Renewal
CGGMP21 (IACR ePrint 2021/060) specifies a UC-secure proactive threshold ECDSA protocol with periodic key-share refresh and identifiable aborts so honest parties can detect and attribute protocol-blocking faults.
Source: IACR ePrint 2021/060 (CGGMP21, updated 2024-10-21), §1 Introduction & Key-Refresh / Identifiable Abort
5e. Rotation & Revocation Across Key Types
Foundational digital-asset key concepts are in Digital Assets (LM-045) and generic KMIP / SP 800-57 key-lifecycle states are in KMS & PQC (LM-016). In custody operations, revocation and rotation behave very differently depending on the layer:
- Ordinary chain keys (Bitcoin UTXO & Ethereum EOA): no revocation exists. A key remains valid for as long as funds sit at its address. “Rotation” means sweeping funds to pre-generated destination keys — which costs fees, takes time, becomes a race after compromise, cannot move locked/staked assets until transferable, and cannot stop future deposits to the old address. Pre-plan destination keys, sweep templates, and quorum-protected emergency-sweep rules.
- Revocable on-chain authorizations: ERC-20 allowances (
approve, EIP-2612permit, Permit2) and EIP-7702 delegations must be revoked on-chain — pausing off-chain withdrawals does not invalidate existing on-chain authorizations. - Multisig & smart accounts: Safe
OwnerManager.soland ERC-4337 smart accounts rotate owners on-chain without moving funds (an account-design property, not an ERC-4337 default). Rule G1: modules, guards, and upgrade keys can bypass owner quorum unless equally constrained. - MPC share refresh is not key rotation: proactive refresh keeps the same public key and address, so it does nothing against a compromised or quantum-exposed key (which still needs a sweep). Refresh security assumes old-epoch shares are erased (§5k); a retained threshold of old shares can still sign.
- Removing a person or device: remove the approver from the quorum (on-chain owner removal, MPC membership-change reshare, or HSM approver-key removal), revoke device/API credentials, and re-verify threshold and separation rules.
- Control-plane keys & descriptors: approver keys, policy-signing keys, TLS/mTLS certs, and HSM wrapping keys rotate via KMS/PKI (CRL per RFC 5280; OCSP per RFC 6960 reports certificate status, not chain key status; platform attestation roots are vendor-controlled). Rotating a wrapping key re-wraps data keys and is the moment to upgrade backup envelopes. Clear-signing descriptors rely on wallet registry freshness and ERC-8176 EAS revocation/expiry because ERC-7730 defines no expiry.
- Incident-response order (Module A1 §3): stop the signing path → revoke credentials and sessions → revoke on-chain allowances/delegations and sweep reachable funds with pre-approved tooling → contact issuers/exchanges for freezes → rotate the control plane.
Bitcoin defines coin ownership as a chain of digital signatures verified against public keys in unspent transaction outputs (UTXOs); the protocol defines no on-chain certificate revocation list for an ordinary UTXO spending key.
Source: Bitcoin Whitepaper (2008-10-31), §2 Transactions
In the Ethereum Yellow Paper, an Externally Owned Account (EOA) address is derived directly from the Keccak-256 hash of its secp256k1 public key, so rotating an EOA key means migrating assets to a new address rather than revoking the old key.
Source: Ethereum Yellow Paper (Shanghai efc5f9a), §4.1 World State & §4.2 The Transaction
Safe Smart Account’s OwnerManager.sol manages a linked list of owner addresses and an m-of-n threshold, permitting authorized calls (addOwnerWithThreshold, removeOwner, swapOwner, changeThreshold) to rotate signers on-chain without moving funds.
Source: Safe Smart Account OwnerManager.sol Source, contracts/base/OwnerManager.sol
Proactive refresh security in threshold schemes such as CGGMP21 relies on honest parties erasing prior-epoch share material once the new epoch is established; retained old-epoch shares can still reconstruct or sign if a threshold of old shares is combined.
Source: IACR ePrint 2021/060 (CGGMP21), Proactive Security Model & Epoch Erasure
EIP-2612 extends ERC-20 with permit() so a signed EIP-712 message sets a token allowance off-chain; once submitted, the allowance remains active on-chain until spent or explicitly revoked.
Source: Ethereum EIP-2612 (Final, 2020-04-13), Specification -> permit()
EIP-7702 writes a persistent delegation indicator (0xef0100 || address) into an EOA’s account code via a type-4 transaction; clearing the delegation requires submitting a new authorization pointing to the zero address.
Source: Ethereum EIP-7702 (Final), Specification -> Delegation Designation
In the 29 October 2026 Binance incident report, Suspending withdrawals and shutting down hot wallet services did not automatically invalidate existing on-chain EIP-7702 authorizations; additional on-chain transactions were needed to revoke them. (Because the MPC hot wallet had no single private key to immediately revoke the malicious EIP-7702 authorizations, Binance developed and executed a specialized MPC procedure and sent a new authorization clearing the EIP-7702 delegation code.).
Source: Binance Security Incident Update & Initial Findings (2026-11-03), EIP-7702 Exploitation & On-Chain Revocation
RFC 5280 defines X.509 v3 certificate path validation and Certificate Revocation Lists (CRLs) for revoking control-plane certificates such as TLS/mTLS and operator credentials.
Source: IETF RFC 5280 (May 2008), §5 CRL and CRL Extensions Profile
RFC 6960 specifies the Online Certificate Status Protocol (OCSP) for querying the current revocation status of an X.509 certificate (certificate status, not on-chain key status).
Source: IETF RFC 6960 (June 2013), §1 Introduction & §2 Protocol Overview
NIST SP 800-57 Part 1 Rev. 5 (final) defines key-lifecycle states, cryptoperiod selection factors, and guidance for key rotation, backup, archival, and destruction.
Source: NIST SP 800-57 Part 1 Rev. 5 (Final, May 2020), §5.3.6 Cryptoperiods & §7 Key Management Lifecycle
NIST SP 800-57 Part 1 Rev. 6 (Initial Public Draft, December 2025 — draft, not final) updates key-management guidance to add FIPS 203/204/205 post-quantum algorithm categories and transition considerations.
Source: NIST SP 800-57 Part 1 Rev. 6 (Draft, 2025-12-05), §5.6 Security Strengths & PQC Key Management
ERC-7730 (Draft) defines a versioned JSON schema for clear-signing descriptors; registry curation, distribution, descriptor expiry, and revocation are outside ERC-7730’s scope and left to wallet registry governance.
Source: Ethereum ERC-7730 (Draft, 2024-02-07), Specification & Security Considerations
ERC-8176 (Draft) defines Ethereum Attestation Service (EAS) attestations for ERC-7730 descriptors, supplying explicit attestation revocation and expirationTime fields.
Source: Ethereum ERC-8176 (Draft, 2026-02-26), Specification -> Attestation Schema & Expiration/Revocation
Quantum view: changing the signature algorithm and moving backup envelopes to post-quantum wrapping are covered in Module B (Custody PQC Migration, optional further reading).
5f. Recovery & Inheritance
Cold-wallet Shamir basics appear in Digital Assets (LM-045) and general M-of-N backup splitting in KMS & PQC (LM-016). Operational recovery matches the mechanism to the failure: lost approver device (§5e), lost MPC share (§5a reshare), lost quorum (pre-built recovery path), site/HSM loss (§5j), or custodian insolvency (client-held recovery material; regulatory asset-return rules are in Module C, optional further reading).
- Mechanisms compared: (1) SLIP-0039 Shamir shares and vendor HSM M-of-N backup cards; (2) vendor client-held MPC recovery kits (check who can run the kit and whether recovery rebuilds the full key in one place); (3) Bitcoin timelocks — relative (BIP-68/BIP-112, capped at 65,535 blocks or ~388 days) vs absolute (BIP-65 for multi-year delays), composed with Miniscript (BIP-379); (4) smart-account guardian modules with delay and veto (Safe{RecoveryHub}, ERC-4337); (5) inactivity (“dead-man”) paths via UTXO churn, smart-account heartbeat, or off-chain custodial process.
- Inheritance (technical patterns only — not legal or estate advice): placing raw secrets in a will that may become public during probate is unsafe. Heirs need asset notice, sealed/independent recovery material or a timelocked heir path, and clear instructions. Avoid single-heir full backups and unrehearsed plans.
- Recovery is an attack surface: attackers target rarely checked recovery paths via social engineering, forged incapacity claims, coercion (Module A1 §2), and the rebuild-in-one-place moment. Mitigate with delays + owner veto alerts, independent guardian channels, scheduled rehearsals, and logging every recovery attempt.
SLIP-0039 specifies Shamir’s Secret-Sharing for Mnemonic Codes, splitting a master secret into M-of-N word-list shares (with optional group thresholds) as a classical alternative to single-copy BIP-39 seed backups.
Source: SatoshiLabs SLIP-0039 (2017-12-18), Abstract & Specification
Thales Luna Backup HSM 7 documentation (vendor pattern) illustrates M-of-N multifactor quorum-authenticated HSM backup and restore to a second HSM within the same cloning domain without exporting plaintext key material to host RAM.
Source: Thales Luna Backup HSM 7 Documentation (Vendor Pattern), Direct Multifactor Quorum Authentication Backup
Fireblocks’ Recovery Utility documentation (vendor pattern) describes client-held Key Recovery Kits encrypted to customer recovery keys so customers can reconstruct assets independently if the custody platform is unavailable — raising the operational question of who can run the kit and whether recovery rebuilds the full key in one place.
Source: Fireblocks Recovery Utility Documentation (Vendor Pattern), Key Recovery Kit & Disaster Recovery
BIP-68 repurposes transaction input sequence numbers for consensus-enforced relative lock-times counted from the confirmation of the output being spent, capped at 65,535 blocks or 65,535 × 512 seconds (approximately 388 days).
Source: Bitcoin BIP-68 (Final, 2015-05-28), Specification -> Relative Lock-Time Encoding
BIP-112 introduces OP_CHECKSEQUENCEVERIFY (CSV) so Bitcoin scripts can enforce BIP-68 relative timelocks on-chain inside spending conditions.
Source: Bitcoin BIP-112 (Final, 2015-08-10), Specification -> OP_CHECKSEQUENCEVERIFY
BIP-65 introduces OP_CHECKLOCKTIMEVERIFY (CLTV) for absolute block-height or Unix-timestamp timelocks, enabling multi-year recovery or inheritance delays beyond BIP-68’s ~388-day relative ceiling.
Source: Bitcoin BIP-65 (Final, 2014-10-01), Specification -> OP_CHECKLOCKTIMEVERIFY
BIP-379 specifies Miniscript for writing composable, analyzable Bitcoin spending policies (such as 2-of-3 now or recovery key after a timelock), while providing no monitoring, cancellation, fee-bumping, or inheritance protocol.
Source: Bitcoin BIP-379 (2023-10-10), Abstract & Specification
Safe{RecoveryHub} documents smart-account recovery modules that let designated recoverers initiate an owner-replacement transaction subject to a review delay window during which current owners can cancel.
Source: Safe Knowledge Base — Safe{RecoveryHub}, Recovery Delay & Cancellation Flow
ERC-4337 decouples transaction validation from ECDSA EOAs through smart-contract accounts, enabling programmable guardian recovery and validation-key replacement as an account-design property rather than a protocol default.
Source: Ethereum ERC-4337 (2021-09-29), Abstract & Account Validation
Quantum view: post-quantum seed and recovery horizons are covered in Module B (Custody PQC Migration, optional further reading).
5g. Key Ceremonies (Procedure, Evidence & Failure Modes)
Digital Assets (LM-045) introduces what a key ceremony is. Here we focus on operational procedure, verifiable evidence, and failure modes — distinguishing physical root-key ceremonies from automated per-vault MPC DKG and from zero-knowledge trusted-setup ceremonies.
- Roles & separation: ceremony administrator (runs script, holds no key material), key custodians/share holders, operators, independent witnesses, and auditor. No single person can both generate and carry away a usable key.
- Preparation & generation: versioned script, tamper-evident bags with recorded serial numbers, verified HSM firmware and hashed offline workstation image, no network, documented entropy (HSM TRNG; physical dice only as a documented self-custody practice). Derive and compare first addresses on a second independent device and test a small transaction and backup restore before real funds arrive.
- Backup, evidence & close-out: seal shares in the room with recorded holder, location, and seal number; retain signed minutes, executed script with deviations, hashes of all output files, witness attestations, and video where policy requires it (auditor/insurer evidence use is in Module C, optional further reading). Sanitize temporary media (§5k), reseal HSMs, and schedule the next rehearsal.
- Distributed remote ceremonies: MPC DKG across sites relies on attested enclaves/HSMs (Module A1 §4), authenticated channels, a shared signed transcript hash, and out-of-band public-key verification.
The ICANN Root Zone KSK Operator DNSSEC Practice Statement (8th Edition) enumerates separated trusted roles for HSM smart-card and safe ceremonies: System Administrators, Crypto Officers, Recovery Key Share Holders, Safe Security Controllers, Internal Witnesses, and Ceremony Administrators.
Source: ICANN Root Zone KSK DPS 8th Edition (2025-04-14), §4.2.2 Trusted Roles
CA/Browser Forum TLS Baseline Requirements v2.2.9 §6.1.1.1 requires for root CAs and externally operated subordinate CAs a key-generation script, a Qualified Auditor witness OR a video recording of the entire generation ceremony, and a Qualified Auditor report (SHALL; SHOULD for internally operated subordinate CAs).
Source: CA/Browser Forum TLS Baseline Requirements v2.2.9, §6.1.1.1 CA Key Pair Generation
NIST SP 800-57 Part 2 Rev. 1 defines split knowledge and dual control as organizational safeguards so no single person can generate, copy, or take away a usable cryptographic key during sensitive lifecycle procedures.
Source: NIST SP 800-57 Part 2 Rev. 1 (May 2019), §2.2 Key Management Policies & Split Knowledge / Dual Control
IANA’s public Root KSK Ceremonies archive publishes executed ceremony scripts, signed minutes, witness attestations, file hashes, and video recordings as a transparent public reference for ceremony evidence.
Source: IANA Root KSK Ceremonies Archive, Published Ceremony Scripts & Evidence Artifacts
CryptoCurrency Security Standard (CCSS) v9.0 covers Key/Seed Generation (1.01) and Wallet Creation (1.02) and requires independently witnessed key ceremonies at Level 3.
Source: CCSS v9.0 (C4, 2024-12-10), Aspect 1: Key/Seed Generation (1.01) & Wallet Creation (1.02)
NIST SP 800-88 Rev. 2 (September 2025) defines media sanitization categories — Clear, Purge (including cryptographic erase), and Destroy — together with verification and documentation of the sanitization outcome.
Source: NIST SP 800-88 Rev. 2 (September 2025), §2.5 Sanitization Methods & §4 Verification and Validation
Zero-knowledge trusted-setup ceremonies (such as the Ethereum KZG Ceremony) are a distinct ceremony class whose sole security goal is destroying every participant’s secret ("toxic waste") rather than preserving a usable custody key.
Source: Ethereum KZG Ceremony Specifications, Overview & Security Model (1-of-N Honesty / Secret Destruction)
Quantum view: dual-algorithm and post-quantum ceremony transitions are covered in Module B (Custody PQC Migration, optional further reading).
5h. Custody Monitoring & Alerting
Tamper-evident HMAC-chained logs and front-running mempool checks are introduced in Digital Assets (LM-045), and general SOC operations in SOC Implementation & PQC (LM-057). Custody monitoring spans four signal classes across three finality stages:
- Four signal classes: (1) signing path (rate anomalies, new destinations, policy-denial bursts, off-hours requests, sub-second blind approvals); (2)key infrastructure (HSM audit events, MPC aborts/unexpected reshares, enclave attestation changes, approver enrollments); (3) governance changes (allowlists, quorums, timelocks, Safe owners/modules/guards, descriptors — always alerted under Rule G1); (4) on-chain signals (outflows, ERC-20
Approvalevents from submittedpermits, EIP-7702 type-4 authorization lists / account code, deposits to retired addresses, look-alike dust). - Three stages, three responses: pre-signing signals can block the transaction; mempool signals give seconds to minutes; confirmed-chain signals cannot undo execution and can only protect later activity (revoke allowances/delegations, sweep, request issuer/exchange freezes).
- Alert design & limits: route high-severity alerts to at least two people on independent channels, alert on monitor heartbeat silence, ship logs off-box to immutable storage, and wire automatic circuit breakers because in real incidents (Module A1 §3) funds moved within minutes. Monitoring detects — it does not prevent.
Azure Cloud HSM’s audit-log tutorial (one vendor product example, not general HSM coverage) logs login, user, policy, wrapping/export, backup, and secure-application operations; each HSM product’s event scope and delivery latency must be checked separately.
Source: Azure Cloud HSM Audit Log Tutorial (2026-10-14, Vendor Pattern), Logged HSM Operation Categories
NIST Cybersecurity Framework (CSF) 2.0 defines the Detect (DE: Continuous Monitoring, Adverse Event Analysis) and Respond (RS: Incident Management, Analysis, Mitigation) functions as a standard vocabulary for custody alerting and playbooks.
Source: NIST CSF 2.0 (CSWP 29, 2024-02-26), Detect (DE) & Respond (RS) Functions
Safe’s ModuleManager.sol emits EnabledModule, DisabledModule, and ChangedModuleGuard events on-chain so monitors can immediately alert on smart-account module authority changes.
Source: Safe Smart Account ModuleManager.sol Source, contracts/base/ModuleManager.sol (Events)
Safe’s GuardManager.sol emits ChangedGuard events whenever a transaction guard hook is attached, replaced, or cleared, providing an on-chain governance-change signal.
Source: Safe Smart Account GuardManager.sol Source, contracts/base/GuardManager.sol (ChangedGuard Event)
The ERC-20 standard requires emitting an Approval event on successful allowance updates (including after an EIP-2612 permit is submitted on-chain), whereas EIP-7702 delegations emit no dedicated event log and require inspecting type-4 transaction authorization lists or account code.
Source: Ethereum ERC-20 (Final, 2015-11-19), Events -> Approval
SEAL 911 (Security Alliance — labeled as one industry emergency-response resource) operates a 24/7 volunteer incident-response channel for coordinating war rooms and issuer/exchange outreach during active crypto exploits.
Source: Security Alliance SEAL 911 (Industry Resource, 2026-11), Incident Response & War-Room Coordination
Tenderly Web3 Actions (labeled as one active vendor pattern) illustrates programmable serverless hooks triggered by on-chain contract events or webhooks to drive custody alerts or automated circuit-breaker pauses.
Source: Tenderly Web3 Actions Documentation (Vendor Pattern), Triggers & Automated Response Hooks
Quantum view: long-term verifiability of signed logs and alert evidence is covered in Module B (optional further reading); incident-reporting deadlines are in Module C (optional further reading).
5i. Access Control & Path Separation
General IAM is covered in IAM & PQC (LM-028) and secrets handling in Secrets Management (LM-022). The custody controls below are operational design patterns (analysis), each with stated residual authority:
- Separate signing and administration paths: signing (request → policy → approval → sign) and administration (policy, allowlist, quorum, keys, firmware, users) use distinct identities, devices, and network routes.
- Separation of duties in the boundary: initiator ≠ approver ≠ policy admin ≠ key custodian ≠ auditor, enforced inside the signing boundary or smart contract, not only in HR policy.
- Human & machine access: phishing-resistant MFA for key roles is a custody design pattern (NIST SP 800-63B-4 makes AAL selection contextual: synced passkeys at AAL2, hardware-bound non-exportable keys at AAL3). Each service uses its own short-lived workload identity (mTLS/token) scoped to one operation and one tier — no long-lived spend API keys.
- Privileged access & limits: just-in-time elevation with second-person approval, recorded admin sessions, dual-controlled break-glass accounts with mandatory alerting (§5h), and joiner/mover/leaver reviews tied to quorum membership (§5e). Access control decides who can ask; it does not stop a deceived or coerced authorized signer.
OASIS PKCS#11 v3.2 separates Security Officer (CKU_SO) and normal user (CKU_USER) sessions and defines CKA_ALWAYS_AUTHENTICATE so sensitive private keys require a C_Login with CKU_CONTEXT_SPECIFIC immediately after C_SignInit.
Source: OASIS PKCS#11 v3.2 (OASIS Standard, 2026-06-03), §4.9 Private Key Objects (CKA_ALWAYS_AUTHENTICATE) & C_Login
The PKCS#11 v3.2 Usage Guide specifies the session-state and user-login access matrix: the SO cannot access private user objects, but on many tokens the SO can initialize or reset the user PIN — an authority assumption Rule G1 must account for.
Source: OASIS PKCS#11 v3.2 Usage Guide (2025-04-15), Session States & User/SO Access Matrix
NIST SP 800-63B-4 makes authenticator assurance level (AAL) selection contextual: it recognizes synced passkeys as phishing-resistant at AAL2 and requires a non-exportable, hardware-bound cryptographic authenticator at AAL3.
Source: NIST SP 800-63B-4 (July 2025), §3.2 AAL2 & §3.3 AAL3 Authenticator Requirements
NIST SP 800-207 (Zero Trust Architecture) frames access control around authenticating and authorizing every request independently regardless of network location.
Source: NIST SP 800-207 (August 2020), §2.1 Tenets of Zero Trust
NIST SP 800-207A extends zero-trust access control to cloud-native multi-location workloads using per-service workload identities (such as mTLS) and granular application-level policy enforcement.
Source: NIST SP 800-207A (September 2023), §3 Workload Identity & Policy Enforcement
NIST SP 800-53 Rev. 5 defines AC-5 (Separation of Duties) and AC-6 (Least Privilege) as control vocabulary for separating payment initiation, approval, policy administration, key custody, and auditing.
Source: NIST SP 800-53 Rev. 5, AC-5 Separation of Duties & AC-6 Least Privilege
Thales Multifactor Quorum Authentication documentation (vendor pattern) illustrates M-of-N smart-card administrative authentication for HSM initialization, backup, and role changes — a vendor capability, not a PKCS#11 standard term.
Source: Thales Luna Multifactor Quorum Authentication Docs (Vendor Pattern), M-of-N Administrative Quorum Overview
Regulatory access-privilege rules are mapped in Module C (optional further reading); FIDO2/WebAuthn post-quantum status is in Module B (optional further reading).
5j. Operational Backup & Disaster Recovery
Operational DR preserves signing capability without turning backups into an easier theft path than the live vault:
- What to back up: key material/shares, safe-usage configuration (policies, allowlists, quorum membership, approver enrolments, timelocks, descriptors), audit/ceremony evidence, and chain-side operating state (derivation indexes, UTXO/account indexes, nonces, pending transactions, customer-to-account mappings — backed up or reconstructable via tested resynchronization).
- Backup authority (Rule G1) & placement: whoever can read, restore, or delete a backup holds signing-equivalent authority. Store across at least two sites in separate risk areas under live-key quorum, keep KEKs separate from backups, alert on every backup access (§5h), and maintain an authoritative inventory against rollback.
- Drills & restore-after-compromise rule: run scheduled restore drills with drill/canary keys (zeroise afterwards); never restore casually onto uncertified hosts. After a compromise, restore from a known-good backup, authenticate and validate later changes before replaying, and reconcile chain state before signing resumes.
NIST SP 800-57 Part 1 Rev. 5 requires that backed-up and archived keys receive at least the same level of cryptographic and physical protection as operational keys, and that key-encrypting keys remain separated from the backed-up keys they protect.
Source: NIST SP 800-57 Part 1 Rev. 5 (Final, May 2020), §6.2.2 Protection of Keys in Storage & §7.3 Post-Operational Phase
NIST SP 800-34 Rev. 1 defines contingency-planning vocabulary including Recovery Time Objective (RTO), Recovery Point Objective (RPO), alternate-site strategies, and scheduled recovery testing.
Source: NIST SP 800-34 Rev. 1 (May 2010), §2.2 Business Impact Analysis (RTO/RPO) & §3.4 Testing
ISO 22301:2019 (paid standard; catalogue identity reference) establishes the business continuity management system (BCMS) vocabulary for operational resilience and disaster recovery governance.
Source: ISO 22301:2019 (Catalogue Reference), Scope & BCMS Structure
AWS CloudHSM documentation (vendor pattern) describes copying encrypted cluster backups across AWS Regions to survive a single-region outage while preserving cluster-level encryption boundaries.
Source: AWS CloudHSM User Guide (Vendor Pattern), Cross-Region Backup Copy
Regulatory business-continuity duties (e.g. DORA) are in Module C §6b (optional further reading); harvest-now-decrypt-later backup re-wrapping is in Module B (optional further reading).
5k. Destruction & Decommissioning
Generic KMIP and SP 800-57 destruction states are in KMS & PQC (LM-016). In custody, this section is the single operational home for retiring chain keys, old-epoch MPC shares (§5e), ceremony media (§5g), and retired backups (§5j):
- Chain keys — destruction can burn funds: follow the sequence sweep → revoke allowances, EIP-7702 delegations, and Safe owner roles (§5e) → announce retirement and monitor address (§5h) → cold retention period → destroy with recorded decision. Destruction never replaces sweeping after a compromise.
- HSMs, KMIP & HD seeds: PKCS#11
C_DestroyObjectremoves one object (refused withCKR_ACTION_PROHIBITEDifCKA_DESTROYABLEis false) and says nothing about clones or backups; KMIP requires Deactivate/Revoke before Destroy; zeroise devices per FIPS 140-3 / ISO/IEC 19790:2012 before disposal or return. For HD seeds, decide at the seed level — destroying a seed destroys every branch derived from it. - Procedural evidence & what NOT to destroy: secret destruction cannot be proved cryptographically; retain witnessed destruction attestations, SP 800-88 Rev. 2 sanitization records, audit logs, ceremony records, and retired public keys/addresses (retention duties are mapped in Module C, optional further reading).
In OASIS PKCS#11 v3.2, C_DestroyObject destroys a single token object and returns CKR_ACTION_PROHIBITED if CKA_DESTROYABLE is CK_FALSE; it makes no guarantees about prior copies, clones, backups, or physical media.
Source: OASIS PKCS#11 v3.2 (OASIS Standard, 2026-06-03), §5.7.3 C_DestroyObject & CKA_DESTROYABLE
In OASIS KMIP v2.1, an Active key cannot be destroyed directly; it must be Revoked or Deactivated first (revoking for compromise transitions it to Compromised), after which Pre-Active, Deactivated, or Compromised keys may be Destroyed.
Source: OASIS KMIP Specification v2.1 (2021-06-20), §4.14 Destroy & §3.6 State Transition Model
NIST FIPS 140-3 Implementation Guidance enforces SSP (Sensitive Security Parameter) zeroisation requirements derived from ISO/IEC 19790:2012 before a cryptographic module is decommissioned or returned.
Source: NIST FIPS 140-3 Implementation Guidance (2026-08-19), Zeroization & SSP Lifecycle Guidance
ISO/IEC 19790:2012 (paid standard; incorporated by reference into FIPS 140-3) specifies cryptographic module security levels and mandatory zeroisation of plaintext sensitive security parameters.
Source: ISO/IEC 19790:2012 (Catalogue Reference), Incorporated by FIPS 140-3 — SSP Zeroisation
Entrust nShield Decommission and Disposal documentation (vendor pattern) details end-of-life HSM zeroisation, tamper-battery/mechanism discharge, and documented physical disposal so un-zeroised hardware is never resold or returned.
Source: Entrust nShield Decommission & Disposal Docs (Vendor Pattern), HSM Zeroisation & End-of-Life Disposal
Quantum view: destroying a classical key does not protect funds behind an already-exposed public key, and retired encrypted backups remain harvest-now-decrypt-later targets until destroyed (Module B, optional further reading).
9A. Lifecycle Control, Evidence & Residual Risk Matrix
| Lifecycle Stage | Primary Controls | Evidence Produced | Residual Risk |
|---|---|---|---|
| 5a. Generation & Refresh | DKG (no single dealer), proactive refresh with old-epoch erasure, identifiable abort | Attested DKG transcript hash, refresh epoch log, erasure attestations | Protocol bugs/complexity (NISTIR 8214); threshold collusion or retained old-epoch shares |
| 5e. Rotation & Revocation | Pre-planned sweeps for EOA/UTXO, on-chain Safe rotation, explicit allowance/EIP-7702 revocation, CRL/OCSP & ERC-8176 for control plane | Sweep tx hashes, zero-allowance / zero-delegation tx receipts, CRL/OCSP & policy rotation logs | Locked/staked assets cannot sweep; race condition after key theft; future deposits to old addresses |
| 5f. Recovery & Inheritance | SLIP-0039 / HSM M-of-N / BIP-65+379 timelocks / Safe guardians with delay + owner veto + independent channels | Rehearsal logs, guardian verification records, timelock/script audit hashes | Social engineering of guardians, coercion, rebuild-in-one-place moment, stale untested shares |
| 5g. Key Ceremonies | Separated roles (admin != custodian), verified firmware/images, independent address check, witnessed/recorded script | Executed script with deviations, signed minutes, artifact hashes, witness/auditor report, media wipe record | Hardware supply-chain implant; unverified RNG; covert recording in ceremony room |
| 5h. Monitoring & Alerting | Signing-path, HSM, governance-change, and on-chain (Approval + EIP-7702 code) monitors with auto-pause & dual-channel paging | Immutable off-box audit logs, heartbeat telemetry, alert-triage & circuit-breaker records | Monitoring detects rather than prevents; confirmed transactions cannot be undone; alert fatigue |
| 5i. Access Control | Separated signing vs admin paths, AC-5 separation of duties, hardware-bound AAL3 MFA, PKCS#11 CKA_ALWAYS_AUTHENTICATE, short-lived workload mTLS | Effective-permission access reviews, JIT elevation logs, recorded admin sessions, break-glass alerts | Deceived or coerced authorized signer (requires quorum, clear signing, and timelocks) |
| 5j. Backup & DR | Cross-site encrypted backups under live-key quorum, separate KEK, authoritative inventory, canary-key drills, known-good restore after compromise | Drill sign-off reports, inventory reconciliation logs, KEK rotation records | Restore cannot recover stolen funds; undiscovered clone or cloud snapshot; rushed restore re-enabling attacker |
| 5k. Destruction & Decommission | Sweep -> unlink -> monitor retention window -> KMIP deactivate -> C_DestroyObject + clones -> FIPS 140-3 zeroisation & SP 800-88 Rev. 2 verification | Witnessed destruction attestations, sanitization verification logs, retained audit logs & retired address inventory | Cannot cryptographically prove no prior copy exists; incomplete backup inventory leaves surviving copies |
Related modules
- Custody Key Threats & ControlsSame track · Industries · Same migration phase · Shares ECDSA, ML-DSA, FIPS 140-3
- Custody PQC MigrationSame track · Industries · Same migration phase · Shares ML-KEM, ML-DSA, ECDSA
- Energy, Utilities & Grid PQCSame track · Industries · Same migration phase · Shares ML-KEM, ML-DSA, ECDSA
- Media, Entertainment & DRM PQCSame track · Industries · Same migration phase · Shares ML-KEM, ML-DSA, ECDSA
Check your understanding
55 questions on Custody Key Lifecycle Operations, 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 Operations Leads, Security Architects & Key Custodians
Capabilities to require in your production tools:
- Pre-planned emergency sweep templates, on-chain ERC-20 allowance and EIP-7702 delegation revocation workflows, and smart-account owner-rotation governance.
- Separated signing and administration paths enforcing AC-5 separation of duties, PKCS#11 CKA_ALWAYS_AUTHENTICATE per-use auth, and short-lived workload mTLS.
- Cross-site quorum-protected backups with authoritative epoch inventories, canary-key DR drills, and FIPS 140-3 / SP 800-88 Rev. 2 verified decommissioning.
How to evaluate in your environment
Execute scheduled canary-key restore drills, ceremony script audits, effective-permission access reviews, and pre-signing vs confirmed-chain alert tabletop exercises in staging.
Production evidence to collect
Signed ceremony minutes and artifact hashes, canary restore drill sign-off, effective-permission review log, immutable off-box alert history, and witnessed sanitization attestations.
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.