1. The incident is about custody, not blockchain failure

On November 5, 2026, the U.S. Attorney’s Office for the Eastern District of Virginia announced that a former FBI agent had pleaded guilty to wire fraud and money laundering after using access to cryptocurrency credentials held in FBI systems. The agreed statement of facts, as summarized by the Department of Justice, says he made about 100 unauthorised transfers connected to 35 investigations and obtained roughly $2.47 million. The plea has been accepted; sentencing is scheduled for March 2027.

The chain did what it was designed to do: it accepted valid signatures and recorded transfers. The failure occurred around authority, credential access and institutional custody. That distinction matters because a transaction can be cryptographically valid while operationally unauthorised and legally wrongful. Blockchain monitoring sees the movement after signing; custody design determines who can create the signature in the first place.

The event should not be used to generalise about a whole agency or to infer facts beyond the public record. It does, however, provide a rare, concrete reason to examine an uncomfortable control question: when investigators collect a private key, seed phrase, hardware device or recovery credential, are they preserving evidence, assuming control of an asset, or both? A system that does not answer that question explicitly creates a powerful insider-risk surface.

DOJ release dated November 5

2. A credential is evidence and an active capability

Traditional evidence can be viewed, measured and copied without immediately changing the thing it describes. A cryptocurrency signing credential is different. It can prove that investigators found access material, but it can also authorize an irreversible transfer. Possession is therefore not merely informational. It may become operational control over value, with consequences for an investigation, victims, counterparties and the integrity of the public ledger.

NISTIR 8387, Digital Evidence Preservation: Considerations for Evidence Handlers, addresses this dual character directly. It notes that digital assets may require special research and safeguards, and that employees with unsupervised access to a password can be difficult to protect from both theft risk and later accusation. The report discusses separate accounts, conversion under appropriate policy and split-key arrangements as possible safeguards. Those are design options, not a universal recipe.

Every intake record should classify the item before anyone attempts a transfer: evidentiary copy, discovery credential, control credential, recovery factor, signing device or already-seized asset. The classification should identify what the item can do, not merely what it looks like. A photograph of a seed phrase and an encrypted backup may be different files but represent the same authority. Treating them as unrelated evidence duplicates the attack surface.

NISTIR 8387 · DOJ Justice Manual

3. Separate preservation from asset control

The first architectural boundary is between proving what was collected and exercising control over funds. Preservation may require a forensic image, cryptographic hash, collection time, source record and restricted original. Asset control may require a transaction, new destination, fee policy, network selection and legal authorization. Combining those steps in one analyst workflow makes it hard to tell whether access was used for examination or movement.

A safer case model creates two linked but distinct objects. The evidence object records provenance, integrity, access and transformations. The custody object records the controlled address or account, authorized purpose, signing policy, destination policy and reconciliation state. An investigator can review evidence without automatically receiving signing capability. A custodian can execute an approved transfer without seeing unrelated sensitive case material.

The link between the objects is itself evidence. It should record how investigators established that a credential corresponded to an address, which network and derivation path were tested, what non-spending verification was used where possible, and who reviewed the conclusion. This avoids a common shortcut: importing a secret into a general wallet merely to see what it controls. Discovery should minimize exposure before operational custody begins.

4. Two-person control must be cryptographic and procedural

A policy that says two people must approve a transfer is weak if one person can still access the complete secret. Strong dual control combines independent authorization with a signing design that prevents unilateral execution. Depending on the asset and legal context, that can involve multisignature accounts, threshold signing, hardware security modules or secret-sharing for recovery. Each choice has different evidentiary, availability and compatibility trade-offs.

Split key is not the same as splitting a seed phrase into two text files. Informal fragmentation can destroy recoverability, expose order and formatting information, or create two copies during assembly. A production design needs a defined threshold, authenticated participants, secure devices, ceremony records, recovery testing and a way to rotate authority when staff or vendors change. No participant should learn more key material than the protocol requires.

Procedural independence matters too. Two approvers on the same team, using the same administrator, device fleet and password-reset channel, may fail together. Approval should bind to a specific transaction intent: network, asset, amount, source, destination, fee ceiling, case identifier and expiry. The signer should reject any payload that differs from the reviewed intent, rather than asking humans to compare opaque hexadecimal data.

5. Minimize and time-bound access

Least privilege for crypto custody is more precise than a generic role such as investigator or administrator. One role may register a case, another may verify an address, another may propose a transaction, and separate custodians may sign. Exporting keys, changing policy, rotating recovery factors and approving emergency movement require their own permissions. Read access to transaction history should not imply access to credentials.

Privileges should be granted just in time and expire automatically. A person assigned to one case does not need standing access to credentials from dozens of investigations. The public DOJ account says credentials connected to 35 investigations were affected, illustrating why broad aggregation can enlarge the blast radius. Segmenting by case, sensitivity, asset type and operational purpose limits how much one compromised identity or malicious insider can reach.

Break-glass access must be real but costly. An emergency path should require a named incident, stronger quorum, independent notification and immediate post-event review. It should not reveal the same master secret kept for normal operations. If emergency access is never rehearsed, the first real incident becomes an experiment; if it is too convenient, it becomes the preferred bypass.

6. Logs must describe authority, not only clicks

NIST defines chain of custody as tracking evidence through collection, safeguarding and analysis, including who handled it, when and why. For cryptocurrency, the record must extend beyond document access. It should capture credential registration, policy changes, signing requests, approvals, device attestations, transaction construction, broadcast, confirmations, reconciliations and any failed or denied attempt.

An audit event should answer five questions: who requested the action, under which case authority, exactly what was approved, which control produced the signature, and what happened on-chain. Logs stored beside the keys are not sufficient; a privileged actor may alter both. Send tamper-evident records to an independently administered system, anchor ordering where useful, and reconcile them against the public chain and custody inventory.

A blockchain transaction is not a complete audit log. It shows that a key authorized a payload, not which employee, approval, device or case purpose was behind the signature. Conversely, an internal approval record does not prove the signed transaction matched the approved intent. The custody platform must join both sides with stable identifiers and retain the unsigned and signed payloads for later verification.

NIST chain-of-custody definition · NIST SP 800-57 Part 2

7. Design monitoring for insider behavior without presuming guilt

Monitoring should look for policy deviations, not profiles of supposedly suspicious people. Useful signals include access outside assignment, repeated viewing of unrelated credentials, export attempts, unusual signing times, destination changes, policy downgrades, failed quorum attempts and movement followed by rapid asset conversion. These are investigation triggers, not proof of intent.

On-chain analytics can add context: first-time counterparties, hops through services, cross-chain movement and conversion patterns may show that funds departed from the authorized custody plan. But familiar graph mistakes still apply. A deposit address can represent a service, a bridge can separate sender from recipient, and a cluster can be probabilistic. Alerts should preserve the raw transaction and explain the rule that fired.

Human review needs independence from the operational chain. The person who maintains the custody system should not be the sole reviewer of its exceptions. High-risk alerts may require security, case leadership and an evidence-management function. Privacy also matters: monitoring must be proportionate, purpose-bound and retained under policy. Insider-risk controls protect both assets and employees when they produce fair, reviewable records.

8. Recovery is part of admissibility and continuity

A perfect control that prevents legitimate recovery can damage a case as surely as theft. Hardware can fail, staff can leave, vendors can disappear and networks can change transaction rules. Recovery plans need tested key shares, successor roles, asset inventories, software and firmware records, network parameters and a procedure for moving authority without losing provenance.

Exercises should cover more than restoration. Can the team reconstruct who approved a historical transfer? Can it prove that a recovered key is the one registered for the case without spending funds? Can it rotate to new custody while preserving old transaction history and legal holds? Can it continue if one location, cloud account or hardware provider is unavailable? Record outcomes and remediate gaps like any other security finding.

Do not put secrets into the exercise report. Document the ceremony, participants, cryptographic identifiers, test transaction policy and evidence produced. A recovery test should demonstrate capability without creating fresh uncontrolled copies. Secure destruction and revocation are also recovery concerns: when authority moves, obsolete shares and devices must be accounted for, disabled where possible and retained only when an evidentiary duty requires it.

9. Build a custody evidence packet for every material transfer

Each material movement should produce a portable packet: legal or case authority, source asset record, destination approval, unsigned transaction, human approvals, signer attestations, signed transaction, network identifier, broadcast time, transaction hash, confirmation policy and final reconciliation. The packet should also state fees, conversions and any difference between expected and received value.

The packet is not a screenshot or vendor dashboard export. It uses machine-readable records with cryptographic hashes and human-readable explanations. It names software versions, data sources and time standards. Sensitive secrets remain excluded. If an outside reviewer cannot access a restricted system, the packet should say what cannot be reproduced and which independent evidence corroborates the event.

This structure supports incident response and ordinary accountability. When a transfer is questioned, reviewers can distinguish an invalid request, a bypassed approval, a compromised signer, a broadcast substitution or a legitimate movement misread by monitoring. Without those layers, every anomaly collapses into the same unhelpful statement: the key signed it.

10. The operating standard is demonstrable control

A custody program should be measured by evidence, not by the existence of a policy document. Useful indicators include the percentage of credentials under cryptographic quorum, privileged access that expires automatically, transactions with complete evidence packets, unreconciled balances, overdue rotations, recovery exercises passed and high-risk exceptions independently reviewed.

Leaders should ask whether one person can discover, assemble and use a credential without another system producing a durable alert. They should ask how many cases share an administrative boundary, how access is removed, what happens when a signer is unavailable and whether every on-chain movement can be mapped to an authorization. The honest answers define the real control environment.

Dor Arad’s central conclusion is that seized cryptocurrency is not merely digital property stored in an evidence room. It is a live authority system inside a live financial network. Preserving it requires the disciplines of forensics, cryptographic key management, treasury operations and insider-risk governance at the same time. The strongest chain of custody is one that can show not only where the evidence went, but why no individual could move the value alone.