Post-Quantum PKI Migration: Planning the Move to ML-DSA Certificates Before the 2031 Signature Deadline
How federal agencies and contractors can plan a public key infrastructure migration to ML-DSA certificates: what NIST IR 8547 says about PKI, parallel and composite hierarchies, certificate sizes, and the order in which to upgrade verifiers and issuers.
Most post-quantum work to date has gone into key establishment, because recorded traffic can be decrypted later. Certificates have had less attention. A forged signature requires a quantum computer at the time of the attack, so the risk is further away.
The schedule for certificates is tighter than that reasoning suggests. A public key infrastructure (PKI) has many parts: certification authorities, hardware security modules, smart cards, enrollment systems, revocation services, and every application that validates a certificate chain. All of them have to accept the new algorithm before the old one can be turned off. This post covers what current guidance says, the design choices involved, and an order of work that avoids outages.
What the guidance says about certificates
NIST finalized IR 8547, Transition to Post-Quantum Cryptography Standards, on October 27, 2026. Table 2 of the report lists RSA, ECDSA, and EdDSA signatures as deprecated after 2030 and disallowed after 2035. It also restates the earlier dates set under Executive Order 14412 for high value assets and high impact systems: post-quantum key establishment by December 31, 2030, and post-quantum digital signatures by December 31, 2031. Our summary of federal PQC mandates covers how those directives fit together.
Three passages in the report apply directly to PKI:
- Section 2.2.4 says PKI components must be updated to issue, distribute, and manage certificates that use post-quantum algorithms, and to sign certificates and revocation information with them.
- Section 3.2 says PKIs using quantum-vulnerable signature algorithms are expected to be deployed and used alongside PKIs using quantum-resistant ones during the transition.
- Section 4.2 names Personal Identity Verification (PIV) as critical trust infrastructure. Because credentials and hardware are replaced on multi-year cycles, the report says updates to cards, readers, middleware, and the supporting PKI have to begin early in the transition period.
The standards needed to build a post-quantum certificate now exist. RFC 9881 (October 2025) defines how ML-DSA keys and signatures are encoded in X.509 certificates and revocation lists. RFC 10073 (November 2026) defines composite signatures that combine ML-DSA with RSA, ECDSA, or EdDSA in a single certificate.
What the Federal PKI is testing
The General Services Administration’s identity program describes its post-quantum readiness work in phases. A first phase tested PIV issuance, authentication, signing, and certificate lifecycle management. A second phase, in progress, aims at a deployable certification authority service blueprint and an operational pilot. The same page describes a working lab with an ML-DSA-87 self-signed root and a bridge certification authority, from which issuing partners can request certificates and test full path discovery and validation.
The draft certificate profiles for that test environment (dated April 21, 2026) show the direction. They describe a parallel post-quantum PKI and use only the “pure” form of ML-DSA. The root certificate worksheet specifies ML-DSA-87. Subscriber certificates may carry ML-DSA-44, ML-DSA-65, or ML-DSA-87 keys, and separate key encapsulation certificates carry ML-KEM keys in place of today’s RSA encryption certificates.
These are test profiles, not production policy. They are still the best public indication of what federal relying parties will be asked to validate.
Parallel hierarchy or composite certificates
A program has two main ways to run old and new algorithms during the transition.
Parallel hierarchies. A second root and set of issuing authorities use ML-DSA only. Subscribers hold two certificates for a period, and each verifier uses the chain it understands. The Federal PKI test profiles follow this model. Each certificate stays simple, but issuance volume doubles for a while, and protocols must be able to select between two credentials.
Composite certificates. One certificate carries both a classical and an ML-DSA key, and a signature is valid only if both components verify. IR 8547 notes that existing NIST standards accommodate dual signatures as long as at least one component algorithm is NIST-approved. It also cautions that hybrid solutions add complexity and typically lead to a second transition to post-quantum-only algorithms. A composite certificate does not help an old verifier, which cannot parse the new algorithm identifier at all.
National security systems have a narrower choice. CNSA 2.0 names ML-DSA-87 as the parameter set for signatures, so a hierarchy that serves those systems needs to be built at that level.
Certificate size is an engineering input
ML-DSA keys and signatures are much larger than the ones they replace. The ML-DSA figures below come from FIPS 204. The classical figures are approximate raw sizes for comparison.
| Algorithm | Public key (bytes) | Signature (bytes) | Key plus signature (bytes) |
|---|---|---|---|
| ECDSA P-256 | 65 | 64 | 129 |
| RSA-2048 | 256 | 256 | 512 |
| ML-DSA-44 | 1,312 | 2,420 | 3,732 |
| ML-DSA-65 | 1,952 | 3,309 | 5,261 |
| ML-DSA-87 | 2,592 | 4,627 | 7,219 |
Every certificate carries one public key and one signature, so the last column is the minimum added per certificate. A leaf and one intermediate at ML-DSA-87 carry about 14 kilobytes of key and signature material before any names or extensions. With ECDSA P-256, the same two certificates carry under 300 bytes.
That difference shows up in smart card storage and read times, authentication over low-bandwidth links, embedded devices with fixed message buffers, and database fields sized for a 2 kilobyte certificate. Measure these in a lab before a profile is approved.
An order of work
1. Inventory verifiers as well as certificates. Certificate discovery tools find issued certificates. The harder list is every place a chain is validated: remote access gateways, servers that accept client certificates, document signing validators, code that pins an issuer, and trust stores built into device images. A cryptographic inventory should record both sides.
2. Upgrade verifiers first. A verifier that understands ML-DSA and still accepts RSA or ECDSA breaks nothing. An issuer that switches before its verifiers are ready causes an outage.
3. Distribute the new trust anchor early. Getting a new root into every trust store is often the slowest step, especially for disconnected systems and fielded devices.
4. Issue in parallel, and cap classical validity. A three-year classical certificate issued in mid-2029 expires in mid-2032, after the signature date for the highest-priority systems. A classical issuing authority created in 2027 with a ten-year life outlasts the 2035 disallowance. Validity periods should be set so that nothing issued from now on has to be revoked early to meet a date.
5. Check enrollment workflows. An ML-KEM key cannot sign anything, so enrollment that relies on a self-signed request to prove possession of the private key needs another method for encryption certificates.
6. Remove classical trust last. Retire the old anchors only after measurement shows that no production validation depends on them.
An example: a device certificate authority
Consider a hypothetical agency PKI that issues certificates to network devices and servers. The details are illustrative.
The root uses RSA-4096. Two issuing authorities use RSA-3072 and issue device certificates valid for three years. Several of the agency’s systems are high impact, so the December 31, 2031 date applies to them.
Working backward, the last classical three-year certificate for a high impact system would have to be issued before the end of 2028 to expire on schedule. The team plans to cut classical validity to one year in 2028 and to have a post-quantum issuing authority in production by then.
Lab testing finds three blockers. The hardware security modules need new firmware to generate ML-DSA-87 keys. One family of older switches cannot store a certificate chain larger than 8 kilobytes. A configuration management tool pins the RSA issuer by key hash in about 600 device templates.
None of these appear in a certificate scan. They are found by tracing where chains are validated and stored, and each has its own lead time.
Keeping track during the overlap
For several years, most organizations will run two hierarchies. The useful questions are specific: which verifiers accept ML-DSA today, which certificates will outlive a deadline, and which devices still present a classical chain after a replacement was issued.
Sentinel helps organizations understand and change cryptography across many systems at once, which includes finding where certificates are issued, validated, and pinned. CAM provides asset-level cryptographic evidence for assets in production after migration, so a program can show per asset which chain is in use. Both products have a provisional patent issued.
If your program is planning a certificate authority migration or setting post-quantum certificate requirements for suppliers, contact QuantumWorks.