Protecting Long-Lived Data From Harvest Now, Decrypt Later Collection
Harvest now, decrypt later risk depends on how long data must stay confidential. How to sort archives and traffic by retention, find exposed key wrapping, and sequence post-quantum fixes.
Most post-quantum planning is organized by system: which servers, which certificates, which appliances. The harvest now, decrypt later (HNDL) threat is organized by data. An adversary who records encrypted traffic or copies encrypted storage today does not care which system produced it. They care how long the contents will still be worth reading. That difference changes what a program office should do first.
NIST’s October 2026 report on the transition to post-quantum cryptography, NIST IR 8547, names the threat directly: adversaries collect encrypted data now to decrypt it once quantum computers are capable. The same publication recommends prioritizing the NIST key establishment standard, FIPS 203 (ML-KEM), for sensitive data that must stay secure for a long time. This post covers how to find that data and what to change about it.
Start with confidentiality lifetime, not system criticality
A system can be critical and still carry data that is worthless in five years. Another can be a low-priority file share that holds personnel records with a retention period of seventy-five years. Ranking by system importance misses the second case.
The usual reasoning tool is Mosca’s inequality. Let X be the number of years the data must remain confidential, Y the number of years your migration will take, and Z the number of years until an attacker can break the current cryptography. If X plus Y is greater than Z, data protected today will be exposed before it stops mattering. Nobody knows Z. X and Y, however, are numbers an organization can estimate from retention schedules and its own project history.
Here is an illustrative sorting for a federal records environment. The retention periods are examples, not requirements.
| Data class | Illustrative confidentiality lifetime | Typical exposure | Priority |
|---|---|---|---|
| Personnel and security clearance records | 50 years or more | Backups, replication links, archive tapes | First |
| Source selection and contract negotiation files | 10 to 25 years | File shares, email, collaboration platforms | First |
| Engineering data and technical data packages | 20 to 40 years | Repository sync, partner exchange | First |
| Routine operational logs | 1 to 3 years | Log forwarding | Later |
| Public-facing web content | None | TLS to public site | Later |
The point of the table is the ordering. The last two rows can wait for the general migration schedule. The first three should not.
Where long-lived data is actually exposed
Teams often say “our archive is encrypted with AES-256” and consider the question closed. That statement describes only one of three places where exposure occurs.
1. Data in transit. A copy of traffic captured on a link is a complete record of every byte that crossed it. If the session key was established with RSA key transport or an elliptic-curve Diffie-Hellman exchange, a future quantum computer can recover the session key from the recording and decrypt the content. This applies to replication between sites, backup transfers, database synchronization, and administrator sessions, not only to web traffic.
2. Stored data whose keys are wrapped with public-key cryptography. Bulk encryption of an archive is usually symmetric. The data encryption key is then protected by a key encryption key, and that wrapping step is sometimes done with an RSA or elliptic-curve key held in a key manager or hardware security module. An adversary with a copy of the ciphertext and the wrapped key has what they need once the public-key layer can be broken. The symmetric algorithm is not the weak point. The wrapping is.
3. Data protected by symmetric keys that travel with it. Long-term symmetric keys with sufficient length are generally considered a lower concern for quantum attack, because known quantum algorithms offer only a limited speedup against them. Their risk is operational: where the keys are stored, how they were first distributed, and who else holds copies. The NSA’s CNSA 2.0 suite specifies AES-256, which is covered in our post on federal PQC mandates.
The first two exposures are the ones a program should look for. The third is usually a key management review.
A worked example: an engineering archive with a forty-year life
Consider a hypothetical program that stores technical data for a platform expected to remain in service for forty years. The data is encrypted at rest with AES-256. Each archive volume has its own data key, and those keys are wrapped with an RSA-2048 key that lives in a hardware security module. Every night, the volumes replicate to a second site over a VPN tunnel that uses an elliptic-curve Diffie-Hellman exchange.
A review of this design finds two HNDL exposures and one non-exposure.
- The nightly replication traffic is recordable. Every volume that crossed the link since the tunnel was built is a candidate for later decryption.
- The wrapped data keys are stored next to the data. An adversary who copies the archive also copies the wrapped keys.
- The AES-256 encryption of the volumes itself needs no change.
The fixes follow from that. The tunnel moves to a post-quantum or hybrid key exchange once the VPN vendor supports it, which closes the exposure going forward but not for traffic already recorded. The wrapped keys are re-wrapped under a key hierarchy that does not rely on RSA, so the stored ciphertext no longer depends on a breakable public-key layer. Re-wrapping a data key is far cheaper than re-encrypting forty years of volumes, which is why it should be done first. Data already captured in transit cannot be protected retroactively. For that material, the decision is whether the contents are still sensitive, and the answer should be written down.
Prefer re-wrapping to re-encrypting, and plan for what cannot be recovered
Three practical rules come out of the example.
First, separate the layers. If the data key hierarchy is clean, changing the outer key protection does not require touching petabytes of stored ciphertext. If keys were derived or distributed in ad hoc ways, you may find that none of the layers can be changed independently. That is itself a finding worth recording.
Second, move key establishment before signatures. A forged signature requires a quantum attack at the time of use, while recorded key exchanges can be attacked at any later date. This sequencing matches the federal direction: NIST describes Executive Order 14412 as requiring federal high-value assets and high-impact systems to migrate key establishment by 2030 and digital signatures by 2031.
Third, accept that the past cannot be fixed. Some data has already crossed networks under classical key exchange. The response is a data-level risk decision: assume it could be read, assess whether that matters, and apply compensating measures such as shortening retention where policy allows.
Questions to ask before rewriting anything
A short list works well in program reviews and in conversations with suppliers.
- What is the retention period of the data, and who set it?
- Which links carry that data between sites, clouds, or partners, and what key exchange do they negotiate?
- Are data encryption keys wrapped with RSA or elliptic-curve keys, and where are the wrapped copies kept?
- Does the key manager or hardware security module support the new algorithms, or does it need a firmware or hardware refresh?
- Does the supplier’s roadmap include FIPS 203 key establishment, and when?
- Can we show evidence of what is deployed today, rather than what the design document says?
The last question matters most. Architecture diagrams routinely describe a tunnel as using one algorithm while a fallback path negotiates another. Our post on building a cryptographic inventory covers how to assemble that evidence from code, configuration, and network observation.
How this connects to QuantumWorks work
Sentinel helps organizations understand and change cryptography across many systems at once. For this problem, that means linking cryptographic assets to the data and retention periods they protect, so that the ordering in a table like the one above comes from evidence instead of estimates. CAM provides asset-level cryptographic evidence for assets in production after migration, so that a re-wrapped key or a changed tunnel can be shown to be in the state the plan says it is. Both products have a provisional patent issued.
If your program holds data that must stay confidential for decades and you want to work through where it is exposed, contact QuantumWorks.