Crypto-shredding makes data unreadable everywhere it exists by encrypting it under a dedicated key and then destroying that key, so backups, replicas and immutable logs that cannot be edited stop holding usable data. It is the most practical way to honour deletion requests and retention limits in systems where finding and overwriting every copy is impossible. This guide explains how to design key granularity, how deletion works in the main cloud key management services, how to handle backups and logs, and where crypto-shredding falls short.
This guide is about deleting your organisation's data on purpose. If you were looking for cryptocurrency fraud, see the Crypto Scam and Drainer Guide. For the wider retention and deletion programme this technique sits inside, see the Data Retention and Secure Deletion Guide. Nothing here is legal advice: confirm with your data protection officer how your regulator views crypto-shredding for your data.
Key takeaways
- Crypto-shredding only works if each unit of data you may need to delete has its own key, or its own data key wrapped by a key you can destroy. Retrofitting it onto data encrypted under one shared key does not work.
- Key destruction is irreversible by design. AWS states that after a KMS key is deleted you can no longer decrypt the data encrypted under it, and Google Cloud states data encrypted with a destroyed key version cannot be decrypted.
- Every cloud key service builds in a delay before destruction: 7 to 30 days in AWS KMS, a 30-day default in Google Cloud KMS, and a 7 to 90-day soft-delete retention in Azure Key Vault. Your deletion timelines must account for it.
- The UK ICO says the key issue for backups is to put data beyond use, even where it cannot be overwritten immediately. Crypto-shredding is a strong way to achieve that, but confirm your regulator's position for your use case.
- Crypto-shredding fails silently if plaintext copies exist anywhere: caches, search indexes, analytics exports, logs written before encryption, or data keys cached in memory or stored unwrapped.
- The keys and the services that can destroy them are high-value non-human identities. A single over-privileged role that can schedule key deletion can wipe data as surely as ransomware.
How crypto-shredding works
The idea is simple. Data is encrypted with a key. If every copy of that key is destroyed, the ciphertext becomes noise, wherever it is stored. You no longer need to find every backup tape, replica, snapshot or append-only log segment that holds the data. You only need to find and destroy the key, which you control in a small number of places.
In practice crypto-shredding is almost always built on envelope encryption (see the Envelope Encryption and KMS Integration Patterns Guide):
- A data encryption key (DEK) encrypts the records themselves.
- A key encryption key (KEK), held in a key management service or HSM, encrypts (wraps) the DEK.
- The wrapped DEK is stored alongside the data. The plaintext DEK exists only briefly in memory when data is read or written.
To shred, you destroy either the KEK (shredding everything it protects) or the wrapped DEK and any other copies of it (shredding one record or one subject). Which one you destroy determines the granularity of deletion, and that design choice has to be made before any data is written.
NIST SP 800-88 Rev. 2, published in September 2025, is the US reference for media sanitisation and lists cryptographic erase among its techniques. It is written for storage media rather than application-level records, but its principle carries across: sanitisation must render access to target data infeasible for a given level of effort, and you need evidence that it was done.
Choosing key granularity
Granularity is the most important design decision. Too coarse and you cannot delete one person's data without deleting everyone's. Too fine and key management cost, latency and operational risk grow quickly.
| Granularity | Typical use | Shred action |
|---|---|---|
| Per tenant | Multi-tenant SaaS, contract termination | Destroy the tenant KEK |
| Per data subject | Erasure requests for personal data | Destroy the subject's DEK and all copies |
| Per retention period | Logs and events with a fixed retention (for example one key per month) | Destroy the key when the period expires |
| Per dataset or system | Decommissioning a product or environment | Destroy the system KEK |
Many organisations combine two levels. A per-tenant KEK in the key management service wraps per-subject DEKs stored in a key table. Erasing one person means deleting their wrapped DEK from the key table and its backups; offboarding a tenant means destroying the tenant KEK. Retention-period keys are the simplest win for logs: rotate to a new key each month and destroy keys as they age out, which enforces retention without touching the log store.
Key per subject has a catch: the key table itself is now the thing you must delete from, and it is backed up too. Keep the key table small, separately backed up with a short retention, and ideally inside the key management service or an HSM rather than in the main database. Otherwise you have moved the backup problem rather than solved it.
Key deletion in cloud key management services
All three major cloud providers deliberately make key destruction slow and reversible for a period, to protect against mistakes and attackers. This section describes each service in alphabetical order, in neutral terms, from its own documentation.
AWS Key Management Service
AWS KMS requires a waiting period of 7 to 30 days when you schedule key deletion, with 30 days as the default, and the actual wait may be up to 24 hours longer. During the waiting period the key is in the Pending deletion state and cannot be used in any cryptographic operation. Deletion can be cancelled with CancelKeyDeletion until the period ends. AWS describes deletion as destructive and irreversible, and states that data encrypted under a deleted key becomes unrecoverable. It also suggests disabling a key instead when you are unsure. Two details matter for crypto-shredding: deleting only imported key material is temporary and reversible if the material is reimported, and for multi-Region keys you must schedule deletion of all replicas before the primary's waiting period starts. AWS logs scheduling and deletion in CloudTrail and supports CloudWatch alarms on attempted use of a key during its waiting period. See the AWS KMS Security Guide for wider controls.
Azure Key Vault
Azure Key Vault soft delete keeps deleted vaults and objects for a retention period of 7 to 90 days, with 90 days as the default. The interval is set when the vault is created and cannot be changed afterwards. Purge protection is optional and not enabled by default; when it is on, a deleted object cannot be purged until the retention period passes. Purging a soft-deleted key requires the purge permission, for example through the Key Vault Purge Operator role. For crypto-shredding, the deletion is final only after purge or after the retention period ends, so record both dates. See the Azure Key Vault Security Guide for wider controls.
Google Cloud Key Management Service
In Google Cloud KMS you destroy key versions, not whole keys. A destroyed version first enters a scheduled for destruction state, with a default duration of 30 days, during which it can be restored. The duration is set when the key is created, and an organisation policy constraint can enforce a minimum. After the scheduled period, destruction is not reversible and data encrypted with that key version cannot be decrypted. The key name and version number remain. Imported key material is the exception: re-importing the same original material restores a previously imported key. See the Cloud KMS and Cloud HSM Guide for wider controls.
What the waiting periods mean for you
Whatever service you use, the data is only provably unrecoverable when the waiting or retention period has ended and the key is gone. If you use imported key material or keep your own copy of key material for escrow, destroying the cloud key is not enough: you must destroy every copy of the material too, or the shred can be undone. The BYOK, HYOK and External Key Management Guide covers where those extra copies usually live.
Erasure requests, backups and immutable logs
A common driver for crypto-shredding is erasure under GDPR Article 17 when personal data sits in backups and logs that cannot be edited. The ICO's guidance on the right to erasure says organisations must respond without undue delay and at the latest within one month, extendable by two further months for complex requests. On backups, it says the key issue is to put backup data beyond use, even if it cannot be immediately overwritten, and that the data within the backup must not be used for any other purpose. The ICO notes that its guidance is under review following the Data (Use and Access) Act.
Crypto-shredding maps well onto that idea. If a subject's data in every backup and log is encrypted with a key that has been destroyed, no one can use it, including you. Practical points:
- Match timelines. A 30-day key waiting period plus a processing queue can push deletion past one month. Start the shred quickly and document the date the key will be irrecoverable.
- Restores must respect shreds. If you restore a database from a backup taken before a subject's key was destroyed, the restored records are unreadable, which is correct. Make sure restore procedures do not also restore the key table from the same point in time, or the shred is undone.
- Immutable storage. Object lock, WORM storage and append-only event logs are good for integrity and ransomware resistance, as the Backup Security and Immutable Backup Guide explains. Crypto-shredding is what lets you keep those protections and still delete.
- Legal holds win. Before shredding, check for litigation holds and statutory retention. Destroying a key under a hold is as serious as deleting the data.
For the end-to-end request process, including identity checks, see the DSAR Guide: Handling Data Subject Requests.
Where crypto-shredding fails
Crypto-shredding only removes data that was encrypted under the destroyed key and nowhere else. Common leaks:
- Plaintext side copies. Search indexes, caches, data warehouse extracts, analytics events, support tickets and machine learning training sets often hold copies that were never encrypted under the subject key.
- Logs written before encryption. Application logs and traces that include personal data in plaintext are not shredded by destroying a database key.
- Cached or exported data keys. Services that cache plaintext DEKs for performance, or that export data keys for client-side use, can hold copies after the shred. Set short cache lifetimes and never persist plaintext DEKs.
- Derived data. Aggregates, hashes and pseudonymous identifiers may still identify a person. Shredding the source does not change what has already been derived.
- Future cryptanalysis. Shredded ciphertext is safe only as long as the algorithm is. Use current authenticated encryption, as described in the Authenticated Encryption and Block Cipher Modes Guide, so that captured ciphertext stays unreadable.
Map every place a data type flows before you rely on crypto-shredding for it. Where a copy cannot be encrypted under the subject key, it needs a conventional deletion path.
Protecting the shredding capability
The ability to destroy keys is the ability to destroy data. Attackers know this: MITRE ATT&CK tracks Data Destruction (T1485), and deleting or scheduling deletion of cloud keys is a quick way to make encrypted data and backups useless. Treat the shredding workflow as privileged:
- Run shredding through one dedicated service identity with permission only to schedule deletion of keys matching your naming or tagging pattern, not all keys in the account or subscription.
- Remove key deletion and purge permissions from general administrator and workload roles. Use organisation-level policies to deny deletion of keys outside the shredding pattern.
- Keep the waiting period at a value that gives your security team time to notice malicious scheduling, and alert on every schedule, cancel and purge event.
- Require a ticket or request reference for each shred and record it with the key identifier, so every destruction can be traced to a lawful reason.
- Back up the key table and keys that are not being shredded, with stronger controls than the data itself.
Proving the data is gone
Regulators, auditors and customers will ask how you know. Build evidence into the workflow:
- A deletion record per request: subject or tenant reference, key identifiers, the date deletion was scheduled, the date the key became irrecoverable, and the identity that performed it.
- Key service audit logs showing the scheduling and final deletion events.
- A periodic test that tries to decrypt a sample of shredded records from a restored backup and confirms failure.
- A data flow map showing which stores are covered by crypto-shredding and which use conventional deletion.
Record retention rules and the shredding method in your policy. The Data Retention and Disposal Policy Template can be adapted to include it.
Common failure patterns
- One key for everything. All customer data is encrypted under one KEK, so the only possible shred deletes the whole system.
- Key table in the main backup. Per-subject keys are stored in the same database and backup set as the data, so a restore brings deleted keys back.
- Forgotten copies of key material. Imported or escrowed key material survives elsewhere, so the shred can be reversed.
- Counting from the wrong date. Teams report deletion on the day they scheduled it, ignoring a 30 or 90-day waiting period.
- Plaintext leaks. Logs, indexes or exports hold the same personal data outside the encrypted store.
- Broad deletion rights. Administrator roles can schedule deletion of any key, creating a fast path to destroy data and backups.
Practitioner checklist
- Decide key granularity (tenant, subject, retention period, system) for each data type before writing data.
- Use envelope encryption with the KEK held in a managed key service or HSM; never persist plaintext DEKs.
- Keep any per-subject key table small, separately backed up with short retention, and excluded from routine data restores.
- Document the waiting or retention period for each key service and include it in erasure response timelines.
- Destroy every copy of imported or escrowed key material when shredding.
- Map plaintext side copies and give each a conventional deletion path.
- Restrict key deletion to one dedicated, narrowly scoped service identity and alert on every schedule, cancel and purge.
- Check for legal holds before every shred.
- Keep a per-request deletion record and test shredded data against restored backups at least annually.
Standards and references
- EUR-Lex: Regulation (EU) 2016/679 (GDPR) (Article 17 right to erasure)
- ICO: Right to erasure
- NIST: SP 800-88 Rev. 2, Guidelines for Media Sanitization
- NIST: SP 800-57 Part 1 Rev. 5, Recommendation for Key Management
- NIST: SP 800-53 Rev. 5 (MP-6 Media Sanitization, SC-12 Cryptographic Key Establishment and Management, SC-28 Protection of Information at Rest)
- AWS: Deleting AWS KMS keys
- Microsoft: Azure Key Vault soft-delete overview
- Google Cloud: Destroy and restore key versions
Related NHI Mgmt Group resources: Envelope Encryption and KMS Integration Patterns Guide · Data Retention and Secure Deletion Guide · Backup Security and Immutable Backup Guide · DSAR Guide: Handling Data Subject Requests