The Algorithm
Is the
Easy Part.
Every post-quantum plan assumes that, when the day comes, you can swap your cryptography. We tested that assumption on our own platform. The algorithm was the least of the problem.
If one of your certificate authorities were compromised tomorrow, could you move everything it issued to a new CA? Could you move it to a new algorithm at the same time?
In October, that stopped being a hypothetical for one CA’s customers. On October 17, Google and Mozilla were told that the private keys of eight intermediate certificate authorities run by eMudhra had been generated with too little randomness. Within three hours of the report, Chrome and Firefox were blocking the ones that issue website certificates. Both then removed the root that signed them, and Google told website operators with certificates under that root that they “must immediately acquire and install a new TLS certificate from another PKI hierarchy”. Chrome set the removal for October 27. Anyone with a website certificate under that root had ten days to move to a different CA.
That is crypto-agility in practice. It is not a property of an algorithm. It is a property of the system that issues, tracks and replaces your certificates. And it is easy to never test, because testing it means pretending to lose a CA.
RSA is not broken. But the margin is shrinking, and AI is part of the reason. In September 2026, researchers at two AI companies used AI coding agents to set two public RSA factoring records 16 days apart: RSA-260 (862 bits), then RSA-896. Neither threatens RSA-2048, the key size most of the internet relies on. Separately, published estimates of the quantum hardware needed to break RSA-2048 fell about tenfold in nine months, from under a million qubits in May 2025 to under 100,000 in February 2026.
The time to prove you can migrate is before you have to. So we pretended to lose a CA.
What We Tested
We issued a set of RSA-2048 certificates from one CA. Then we changed one thing, the issuing policy, pointing it at a new CA and a new algorithm, and renewed every certificate under it. We ran that migration three times.
- ECDSA P-256. The modern classical default.
- ML-DSA-65. NIST’s post-quantum signature standard (FIPS 204) at its middle security level.
- ML-DSA-87. The same standard at its highest level.
The run was automated end to end: check preconditions, issue, switch the policy, migrate, verify, then restore. No hand-edited certificates, no manual steps in the middle.
What We Found
All three migrations worked. Every certificate we sampled came back from the new CA with the new key type and the new signature algorithm, and the old certificate was marked superseded.
We did not take our own software’s word for it. We pulled every sampled certificate and parsed it independently with OpenSSL. The key, the signature and the issuer all matched what the policy asked for.
Why the Algorithm Is the Easy Part
Post-quantum migration is usually framed as a cryptography problem. Having run one, we think it is mostly a systems problem. Four things mattered more than the choice of algorithm.
- Policy is the control point. The migration worked because each certificate was bound to a policy, not to a hand-picked algorithm. Change the policy and everything under it follows. Certificates issued outside policy are the ones you will not be able to move.
- The pipeline sets the pace, not the signature. Signing is fast; we measured that in FN-09. The time goes into everything around it: generating and storing keys, recording state, coordinating thousands of renewals at once. Size a migration there.
- Verify outside the tool. A dashboard shows what the system intended. Only the certificate shows what actually happened. Every migration plan should end with independent verification of what was issued.
- Re-issuance is half of recovery. A new certificate that has not reached the load balancer, the service or the device protects nothing. Real recovery is re-issuance, plus deployment, plus revoking the old certificates. Measure all three.
What Comes Next
We are running this again at production scale, with deployment to managed systems and revocation counted in the clock. We are also building migration into the product as a standard operation, so it is not something a team has to script in the middle of an incident.
Four Questions for Your Team
- Do you know which of your certificates were issued outside a policy?
- If a CA were compromised today, what is the first step, and who takes it?
- Has anyone checked, outside your tooling, which algorithm your certificates actually use?
- Does your plan end at re-issuance, or at the last endpoint serving the new certificate?
If any of those took more than a few seconds to answer, that is your crypto-agility gap. It is not the algorithm.
Field Note FN-15-2026 · Distributed under arkion.ai/field-notes
For questions or to discuss findings: research@arkion.ai
- Arkion, crypto-agility test, October 31 to November 1, 2026.
- Google, Chrome and Android Security Teams, “Protecting Chrome and Android Users from Compromised eMudhra Intermediate CA Certificates”, October 20, 2026.
- Mozilla, Firefox Security Team, “Protecting Firefox users following the compromise of eMudhra intermediate certificates”, October 20, 2026.
- Cisco Talos, “Compromised eMudhra intermediate certificates”, October 21, 2026.
- Eric Lu, Cognition, “Factoring RSA-260”, September 9, 2026.
- Stephen A. Weis, “RSA-896”, September 2026 (factored September 19, 2026).
- Craig Gidney, “How to factor 2048 bit RSA integers with less than a million noisy qubits”, arXiv:2505.15917, May 2025.
- Paul Webster et al., “The Pinnacle Architecture: Reducing the cost of breaking RSA-2048 to 100 000 physical qubits using quantum LDPC codes”, arXiv:2602.11457, February 2026.
- Cloud Security Alliance, “Harvest Now, Decrypt Later: Enterprise PQC Migration Gap”, May 31, 2026.
- NIST, FIPS 204, “Module-Lattice-Based Digital Signature Standard”, August 2024.
Recommend Arkion in one click.
Prove you can migrate
before you have to.
In ArkionID, every certificate is issued under a policy you can change. Book a demo and we will show you how issuing policies and certificate authorities work together.