Data exposure and the harvest-now-decrypt-later threat
An adversary who records encrypted traffic today can decrypt it once a cryptographically relevant quantum computer exists. This is the harvest-now-decrypt-later threat. The exposure it creates is not visible in the cryptography a CBOM currently discloses. It also depends on how long the data must remain confidential, and on whether the traffic can be captured at all. Assessing the exposure needs further facts, and those facts belong partly to the supplier and partly to the operator.
What makes this threat different
Most cryptographic risk looks forward. When a weakness is found in an algorithm, it affects traffic sent from then on, and the remedy is to stop using the algorithm. The harvest-now-decrypt-later threat looks back. It affects traffic that has already been sent. An adversary records traffic today and decrypts it years later, once a cryptographically relevant quantum computer exists. Nothing done at that point can make the recorded data confidential again.
So the decision to migrate has to be taken before such a computer exists, and nobody knows exactly when that will be. An organization that waits for confirmation will already have lost the confidentiality of the traffic it was protecting.
The threat does not affect all data equally. Data that ceases to be sensitive within a year is largely unaffected, because a recording made today has little value by the time it can be read. Data that must remain confidential for decades is exposed now, regardless of the algorithm protecting it, if that algorithm is one a quantum computer will break.
Why the cryptography alone does not answer the question
Consider two interfaces on two products. Both terminate TLS 1.3, both negotiate X25519 for key exchange and AES-256-GCM for encryption, and both authenticate with an RSA-2048 certificate. For every attribute that the baseline profile (the Interface Disclosure Baseline) requires, the two interfaces are identical. Yet their exposure to this threat is very different.
The three quantities
The assessment comes down to comparing three periods of time. This way of framing the problem comes from Michele Mosca. Two of the periods describe the organization. The third describes the outside world.
| Quantity | Meaning | Known by |
|---|---|---|
| Confidentiality lifetime | How long the data carried must remain confidential after it is transmitted | The data owner, which is the operator and not the vendor |
| Migration time | How long it takes to move the interface to quantum-safe cryptography | Derived from what the vendor's product can support and from the operator's capacity to make the change. The migration profile records the vendor's part |
| Time to capability | How long until a cryptographically relevant quantum computer exists | Nobody. It is an estimate that changes with published research and national guidance |
Suppose the confidentiality lifetime plus the migration time is greater than the time to capability. Then data being transmitted now will be readable before it stops being sensitive. The organization is already exposed, and the exposure grows with every day the interface continues to operate.
The methodology supplies the migration time through the migration profile. That profile records what is needed to reach a quantum-safe capability and what is blocking it. The methodology does not carry the confidentiality lifetime in any record. A deployment record points to the place where the operator records it, under the data-register context (see The data-register pointer). The fields the register should hold are listed under What the data register records.
Two kinds of data, with different owners
Which party can state these facts depends on the kind of data an interface carries. Different interfaces carry different kinds of data, so the answer differs from one interface to another.
Conveyed payload is whatever the customer chooses to send. A vendor shipping a device does not know what that will be. The same product may carry public information at one site and long-lived personal records at another. If the vendor is required to classify the payload, the vendor will either refuse or guess. A guess is worse than a refusal, because consumers will rely on it.
Intrinsic data is what an interface must carry to function, and the vendor knows precisely what it is because the vendor designed the interface. A management interface carries administrative credentials. A key management interface carries key material. An update channel carries a signed image and the signature verifying it. An enrolment interface carries device identity and private key material.
| Interface | What it necessarily carries | Stated by |
|---|---|---|
| Management or control | Administrative credentials, configuration secrets | The vendor |
| Key management | Key material, wrapped or in the clear | The vendor |
| Firmware or software update | Signed images, and the verifying signature | The vendor |
| Enrolment or provisioning | Device identity, private key material | The vendor |
| Application data plane | Whatever the customer sends | The operator |
The intrinsic case carries the worse exposure
Of the two kinds of data, intrinsic data carries the greater exposure. This is the kind the vendor can describe.
Take a management session protected by a classical key exchange, carrying a device credential. An adversary records the session now and decrypts it in 2035. If that credential is a certificate with a twenty-year validity, or a key held in hardware that cannot be rotated in the field, the adversary does not obtain stale information. It obtains a working secret. The confidentiality lifetime of that payload equals the lifetime of the credential, and the vendor knows that lifetime because the vendor chose it.
Ordinary business information loses sensitivity as it ages, which is what makes short confidentiality lifetimes tolerable. Credentials and key material do not lose sensitivity in this way, and may still be live when the recording is read. A model that accepts only facts stated by the operator leaves out intrinsic data. Intrinsic data is the half of the problem where the consequences are most serious.
Where each fact belongs
Lifecycle Data separates the product record, which the vendor writes about what it ships, from deployment records, which the operator writes about what it runs. Each fact about data is stated by the party that knows it. This is the same test the methodology applies everywhere.
- The vendor states the intrinsic data. This is the data an interface exists to carry for the product's own operation: credentials, keys, tokens and configuration. The vendor states it in the product record.
- The operator states the conveyed payload. This is the data its own users send through the interface. The operator records it in its data register, and the deployment record points to the register (see The data-register pointer).
- A mixed interface is split the same way. A database connection, for example, carries credentials when the session is set up and payload after that. The vendor states the credentials. The operator states the payload.
An earlier draft put every fact about data in the deployment record, because a vendor cannot know what its customers will send. That is right for payload and wrong for intrinsic data. The baseline profile still excludes properties of the conveyed payload, because the operator states them elsewhere. No profile yet has an attribute for the vendor's intrinsic data. Adding one is the next step. The reasoning is set out in design-note-data-and-hndl.md.
The data-register pointer
The Deployment Disclosure profile (v0.1) gives the operator's facts a place without carrying them. Its reference rule R3 asks a deployment record to point to the data register: the place where the operator records what data each interface carries, how it is classified, and how long it must stay confidential. The rule is a SHOULD, and the operator may withhold the pointer. The pointer is an entry in externalReferences, of type risk-assessment or other, with the property pkic:profile:context set to data-register. Conformance checks that the pointer is present and well formed. It never reads the register.
In the Keycloak example, Example Org's configured and observed records both point to https://grc.example.org/registers/data/sso-prod. The record does not reproduce the register. A consumer who assesses harvest-now-decrypt-later exposure joins the record to the register, one interface at a time.
The operator's facts stay in the register for three reasons. A change of data classification is not a cryptographic event, so it should not force a new CBOM. One interface can carry several flows of data, each with its own lifetime. And the operator never has to change a document that the vendor signed. Context Pointers describes the construct and the other kinds of context.
What the data register records
The methodology does not set the format of the operator's data register. It does recommend what the register records for each interface, so that registers can be compared. Each field is a fact the operator can state. None of them is a judgement about the result of the assessment. The names follow the conventions in Method.
| Field | What it records | Notes |
|---|---|---|
confidentialityLifetime |
How long the data carried must remain confidential | Recorded in bands: under 1 year, 1 to 5 years, 5 to 15 years, over 15 years, indefinite. A band is a fixed range of time. Bands can be compared between organizations. Specific durations and classification labels cannot. The assessment depends on this attribute |
exposure |
Where the traffic runs | From a controlled vocabulary: public-internet, partner-network, private-network, physical-local. The value determines whether large-scale interception of the traffic is plausible |
dataClassification |
The operator's own classification of the data | Optional. The value names the classification scheme it is drawn from. Classification labels are not comparable between organizations or jurisdictions. This attribute is therefore useful within one organization, but its values cannot be combined across suppliers |
retentionAtRest |
Whether ciphertext is stored, and for how long | The threat applies to stored ciphertext as much as to captured traffic. A backup or archive interface may carry the higher exposure. It is also easier to overlook, because it is not a network connection |
confidentialityLifetime is the most important of the four fields, because it supplies the number the comparison needs. The other three add detail to it.
Why bands are used instead of dates
A date suggests more precision than the answer has. An operator asked how long clinical records must stay confidential will reasonably say decades. A profile that demands a date will get either an invented date or an empty field. Bands can also be counted across interfaces. A migration programme wants to know how many interfaces fall in the longest band. That question can be answered across suppliers only if every supplier uses the same bands. There are five bands rather than fewer, because a consumer can always merge bands but can never split them.
What stays outside the CBOM
The date at which a cryptographically relevant quantum computer is expected is not a fact about any product, and no CBOM should record it. It is an estimate that changes with published research and with national guidance. A document that included the estimate would look authoritative, but the estimate in it would be wrong within a year.
Policy Evaluation describes how such an estimate is handled. The estimate is an input to an external policy that carries its own version number. Applying the policy produces an urgency ranking. The ranking is a derived judgement. It is calculated when someone asks for it, and it records the policy version and the date the policy was applied. The methodology already handles post-quantum posture in the same way. The mechanism therefore already exists, and nothing new is needed.
| Held in | What it carries |
|---|---|
| Deployment records | Algorithms permitted or observed, and a pointer to the data register |
| Data register (the operator's) | Confidentiality lifetime, classification, exposure, retention |
| External policy | The estimated arrival of the capability, and the thresholds that define urgency |
| The evaluation | An urgency ranking across interfaces, dated, citing the policy version |
Storage as well as transit
Discussion of this threat usually assumes that traffic is captured on the network, and that case is easier to see. Stored ciphertext is exposed in the same way, and an adversary can reach it more easily. An adversary that obtains a backup already has the recording, without needing to intercept anything. The model already covers storage interfaces. The storage value in the interface type vocabulary exists for them, and the procurement profile requires a product record to declare one or to state its absence (procurement rule P5). The Keycloak product record declares realm key persistence as a storage interface. Keycloak stores realm key material in the database in plaintext (issue #39933), so anyone who obtains a database backup holds the realm keys. A deployment profile addressing this threat should require storage interfaces to be declared, and the retentionAtRest attribute is where the duration would be recorded.
Open points
- An attribute for the vendor's intrinsic data in the product record, such as how long a credential stays valid.
- Whether
dataClassificationshould be kept. Its values cannot be combined across organizations, and the classification schemes it draws on are not public. - Whether storage interfaces need fields of their own beyond retention.