PCI DSS key management: requirements 3.6 and 3.7 without an HSM team
By Sam Rivera, Founder, SentinelPanda · October 24, 2026 · 3 min read · PCI DSS
Encrypting the PAN moves the problem to the key. Requirements 3.6 and 3.7 are about that problem.
When this applies
Requirements 3.6 and 3.7 apply when you use encryption to render stored PAN unreadable under 3.5.1. They do not bite in the same way if you use truncation, or if a third-party tokenisation provider holds the keys and you hold only tokens — in that case the key management sits with the provider and shows up in their attestation of compliance.
If you do hold the keys, these two requirements are often where an otherwise clean requirement 3 fails, because the encryption is solid and the paperwork around the keys is missing.
Requirement 3.6: protecting the keys
- Access to keys is restricted to the fewest custodians necessary (3.6.1).
- Key-encrypting keys are at least as strong as the data-encrypting keys they protect, and are stored separately from them.
- Keys are stored in the fewest possible locations (3.6.1.4).
- Secret and private keys are stored in one of three forms: encrypted with a key-encrypting key, inside a secure cryptographic device such as an HSM, or as at least two full-length key components or key shares (3.6.1.2).
- Service providers additionally maintain a documented description of their cryptographic architecture (3.6.1.1).
Requirement 3.7: the lifecycle
- Strong key generation (3.7.1), secure distribution (3.7.2) and secure storage (3.7.3).
- Key changes at the end of a defined cryptoperiod (3.7.4) — you define it, and you have to be able to show it.
- Retirement or replacement when a key's integrity is weakened, including when someone with knowledge of a cleartext component leaves, or when compromise is suspected (3.7.5).
- Split knowledge and dual control for any manual cleartext key-management operation (3.7.6).
- Prevention of unauthorised key substitution (3.7.7).
- Key custodians formally acknowledge their responsibilities, in writing or electronically (3.7.8).
What a cloud KMS changes
AWS KMS, Azure Key Vault and Google Cloud KMS generate and hold keys in hardware security modules and never release root key material in cleartext. Envelope encryption — a KMS key wrapping the data keys your application uses — gives you the key-encrypting and data-encrypting separation that 3.6.1 describes, and scheduled rotation covers much of 3.7.4.
What the platform cannot decide for you is the cryptoperiod, who holds administrative and decrypt rights, the custodian acknowledgements, and the procedure for suspected compromise. That procedure has to include re-encrypting existing data, not just rotating the key going forward.
The decrypt permission is the real boundary
In a KMS design, whoever can call decrypt on the key can read the PAN. That makes the key policy and the IAM roles that reference it the most important access control in requirement 3, and it puts every principal with that permission in scope. Keep it to the application role that genuinely needs it, log every call under requirement 10, and treat any break-glass access as a key-management event.
Evidence assessors ask for
- A key inventory: purpose, algorithm and strength, location, custodian, cryptoperiod and last rotation for each key.
- Signed custodian acknowledgements for everyone with key-administration rights.
- KMS key policies and the IAM roles allowed to use them.
- Rotation records and the written procedure for suspected compromise.