# How certificate authorities generate keys, and how it failed

> How a CA should generate private keys (script, HSM, witnesses, key checks), how eMudhra's weak P-384 keys got through in October 2026, and what site owners do.

Canonical: https://red.run.cloud/learn/how-certificate-authorities-generate-keys · By Timothy Lindblom · Published 2026-11-02 · Updated 2026-11-02

A certificate authority (CA) should generate its private keys inside a hardware security module, in a documented and witnessed ceremony, with a random number source and a check of the result. In October 2026 eight eMudhra intermediate CA keys were found to be tiny numbers, recoverable almost instantly. Replace any certificate that chains to them and update trust stores.

| Fact | Detail | Source |
|---|---|---|
| What failed | Key generation. Eight P-384 intermediate CA private keys with values below 256 | [CERT-In, 21 Oct 2026](https://www.cert-in.org.in/s2cMainServlet?CACODE=CICA-2026-3649&pageid=PUBADV01) |
| Scope | Eight intermediates: six TLS, two code signing | [Mozilla Bugzilla 2083449](https://bugzilla.mozilla.org/show_bug.cgi?id=2083449) |
| Reported to Mozilla | 17 October 2026, 20:06 UTC. Mozilla recovered the keys itself at 21:48 UTC | Bugzilla 2083449 |
| Firefox block | OneCRL update 17 October 2026, 22:59 UTC | Bugzilla 2083449 |
| Revoked by the CA | 18 October 2026 | Bugzilla 2083449 |
| Mozilla root trust | Trust removed for emSign ECC Root CA - G3 on 20 October 2026 | Bugzilla 2083449 |
| Chrome Root Store | Removal of emSign ECC Root CA - G3 set to reach devices on the morning of 27 October 2026 (PDT) | [Google](https://blog.google/security/protecting-chrome-and-android-users-from-compromised-emudhra-intermediate-ca-certificates/) |
| The CA's response | eMudhra said on 22 October 2026 it is leaving the global SSL/TLS certificate business and replacing affected certificates free through partners | [CSO Online](https://www.csoonline.com/article/4239634/emudhra-quits-global-tls-certificate-market-after-compromise-of-ca-signing-keys-2.html) |
| Malicious use | None confirmed in the sources we reviewed | CERT-In, Bugzilla |
| Checked | 2 November 2026 | this page |

## How are CA private keys supposed to be generated?

A CA key is the thing that makes every certificate under it believable. So the rules for creating one are stricter than for any server key. Four controls matter, and each maps to a published requirement.

1. **A written script, followed in a ceremony.** The CA/Browser Forum [Baseline Requirements](https://cabforum.org/working-groups/server/baseline-requirements/requirements/) (version 2.3.1, effective 4 October 2026) section 6.1.1.1 asks for a documented generation procedure, a witness in person or on recorded video, and an auditor who checks what happened against the documented procedure.
2. **A hardware security module (HSM).** The key is created inside tamper-resistant hardware and is not meant to leave it in clear form. The Forum's [Network and Certificate System Security Requirements](https://cabforum.org/working-groups/netsec/requirements/) cover the systems around that key.
3. **A proper random source.** NIST describes approved random bit generators in [SP 800-90A](https://csrc.nist.gov/pubs/sp/800/90/a/r1/final) and key management in [SP 800-57](https://csrc.nist.gov/pubs/sp/800/57/pt1/r5/final). The signature algorithms themselves are in [FIPS 186-5](https://csrc.nist.gov/pubs/fips/186-5/final).
4. **A check of the result.** Baseline Requirements section 6.1.6 expects public key parameters to be verified as not weak. Section 6.1.5 limits ECC keys to P-256, P-384 and P-521.

Browser programs add an audit layer. The [Mozilla Root Store Policy](https://www.mozilla.org/en-US/about/governance/policies/security-group/certs/policy/) (version 3.1, effective 1 July 2026) asks for an auditor-witnessed root key generation ceremony report and continuous annual audits afterwards.

<figure class="ln-figure">
<svg viewBox="0 0 720 250" role="img" aria-labelledby="ln-ca-flow-title" xmlns="http://www.w3.org/2000/svg" font-family="Geist, sans-serif">
<title id="ln-ca-flow-title">The CA key generation path and where the October 2026 incident broke it</title>
<rect x="0" y="0" width="720" height="250" fill="#FFFFFF"/>
<g font-size="14" fill="#1A0F0F">
<rect x="14" y="50" width="120" height="70" rx="10" fill="#FFF1EF" stroke="#EAD3CF"/>
<text x="74" y="80" text-anchor="middle">Written</text><text x="74" y="100" text-anchor="middle">script</text>
<rect x="156" y="50" width="120" height="70" rx="10" fill="#FFF1EF" stroke="#EAD3CF"/>
<text x="216" y="80" text-anchor="middle">Ceremony in</text><text x="216" y="100" text-anchor="middle">an HSM</text>
<rect x="298" y="50" width="120" height="70" rx="10" fill="#FFF1EF" stroke="#EAD3CF"/>
<text x="358" y="80" text-anchor="middle">Witness and</text><text x="358" y="100" text-anchor="middle">auditor report</text>
<rect x="440" y="50" width="120" height="70" rx="10" fill="#FFF1EF" stroke="#EAD3CF"/>
<text x="500" y="80" text-anchor="middle">Public key</text><text x="500" y="100" text-anchor="middle">quality check</text>
<rect x="582" y="50" width="124" height="70" rx="10" fill="#FFF1EF" stroke="#EAD3CF"/>
<text x="644" y="80" text-anchor="middle">Certificate</text><text x="644" y="100" text-anchor="middle">signed</text>
</g>
<g stroke="#655553" stroke-width="2"><line x1="134" y1="85" x2="156" y2="85"/><line x1="276" y1="85" x2="298" y2="85"/><line x1="418" y1="85" x2="440" y2="85"/><line x1="560" y1="85" x2="582" y2="85"/></g>
<rect x="156" y="150" width="262" height="64" rx="10" fill="#C70500"/>
<g font-size="14" fill="#FFFFFF"><text x="287" y="177" text-anchor="middle">Second, weak key pair created</text><text x="287" y="197" text-anchor="middle">and picked by mistake (CA account)</text></g>
<line x1="216" y1="120" x2="216" y2="150" stroke="#C70500" stroke-width="2"/>
<g font-size="13" fill="#655553"><text x="440" y="176">A check of key values at this step</text><text x="440" y="196">would have flagged values below 256</text></g>
</svg>
<figcaption>The controls a CA is expected to run, and where eMudhra says its ceremony went wrong. The last note is our reading of the Baseline Requirements, not a finding in the incident report.</figcaption>
</figure>

## What actually went wrong at eMudhra?

On 17 October 2026 a researcher sent proof of control of private keys for eight eMudhra intermediate certificates to the browser programs and the CA. In [Mozilla's incident record](https://bugzilla.mozilla.org/show_bug.cgi?id=2083449), the P-384 private keys all used integers below 256, "effectively all-zeroes apart from the last byte". Mozilla recovered the keys independently within hours.

[CERT-In's advisory](https://www.cert-in.org.in/s2cMainServlet?CACODE=CICA-2026-3649&pageid=PUBADV01) says the P-384 private keys "had values smaller than 256 and could be recovered almost instantly from their corresponding public keys". Small numbers in write-ups are key values, not a count of certificates. There were eight intermediates. A shorter write-up for site owners is in [the eMudhra weak keys explainer](/learn/emudhra-weak-keys-explained).

eMudhra's explanation, quoted in both sources, is a flaw in the key generation ceremony of 30 April 2025. A wrap and unwrap test step produced a second set of weak keys alongside the intended strong ones, and the weak pair was used by mistake when creating the signing requests, because the key labels were similar. That is the CA's own account. Mozilla sent the CA written questions on 20 October, and they were overdue by 23 October. The Chrome Root Program also posted questions and called the CA's first response inadequate. eMudhra posted its answers to both programs in the [Bugzilla thread](https://bugzilla.mozilla.org/show_bug.cgi?id=2083449) in the last days of October, and at our 2 November check the full incident report for this bug was still outstanding.

## Does a weak key mean the algorithm was broken?

No. P-384 is an approved curve. A private key is supposed to be a number chosen at random from a range of about 2^384 values. If it is chosen from fewer than 256, anyone can try every candidate and match it to the public key. The same would be true of an RSA key built from predictable primes.

That is why "weak RSA key" and "weak ECC key" incidents read alike. This one used ECC. Earlier cases show the same pattern with different parts failing:

| Case | What was wrong with key generation | Source |
|---|---|---|
| Debian OpenSSL, 2008 | A packaging change left the random number generator predictable, so keys came from a small set | [DSA-1571-1](https://lists.debian.org/debian-security-announce/2008/msg00152.html) |
| ROCA, 2017 | A vendor library built RSA primes with a structure that allowed factoring from the public key | [CCS 2017 paper](https://crocs.fi.muni.cz/public/papers/rsa_ccs17) |
| eMudhra, 2026 | A script flaw created a second set of ECC keys with values below 256 | [Bugzilla 2083449](https://bugzilla.mozilla.org/show_bug.cgi?id=2083449) |

In each case the standard algorithm was fine. The failure was in producing, selecting or checking the key. Whether algorithm choice matters for a small site is a separate question, covered in [post-quantum TLS for small sites](/learn/post-quantum-tls-does-a-small-site-need-to-act).

## Why did the controls not stop it?

The reviewed sources do not give a full answer, and the CA owes one. Mozilla's record lists open questions to the CA, and at our 2 November check eMudhra had not filed its full incident report there. We will not guess at which control should have caught it.

What we can say from the requirements is narrower. A ceremony witness and an auditor confirm that the script ran as written. If the script itself carried the flaw, a witness following it would see a normal-looking run. A separate technical check of the key values is what catches that class of error, which is the point of section 6.1.6.

## What changed in browsers and operating systems?

Revocation did not reach every client at the same time, so trust stores are the main thing to update.

- **Firefox:** blocked the six TLS intermediates through OneCRL on 17 October. Per [Mozilla's blog](https://blog.mozilla.org/security/2026/10/20/protecting-firefox-users-following-the-compromise-of-emudhra-intermediate-certificates/), Firefox 158.0.1, ESR 115.43.1 and ESR 153.5.1 cover users whose settings blocked OneCRL delivery, and NSS 3.131.1 removes the root that signed them.
- **Chrome:** [Google](https://blog.google/security/protecting-chrome-and-android-users-from-compromised-emudhra-intermediate-ca-certificates/) blocked the affected intermediates through a CRLSet update on 17 October at 19:45 UTC, pushed the same block to Android through Google Play Services on 19 October, and set the removal of emSign ECC Root CA - G3 from the Chrome Root Store to reach devices on the morning of 27 October 2026 (PDT).
- **RHEL:** Red Hat's [RHSB-2026-012](https://access.redhat.com/security/vulnerabilities/RHSB-2026-012) lists updated `ca-certificates` errata for RHEL 8, 9 and 10 and blocklist guidance. It was last updated 23 October 2026.
- **GitLab:** the [29 October patch release](https://docs.gitlab.com/releases/patches/patch-release-gitlab-19-5-3-released/) (19.5.3, 19.4.5 and the other lines) removes trust for the eight intermediates as CVE-2026-111943. Runner checks certificates itself, so update it to 19.5.1 and rebuild job images that ship their own `ca-certificates`.
- **Your own stacks:** language runtimes, Java stores, containers and anything that bundles a CA list. Check each vendor's advisory and rebuild images from updated bases.

## Am I affected? A three-state self-test

<div class="ln-selftest">
<h3>Am I affected? Where you stand right now</h3>
<p>From outside, a check can read the certificate chain your site serves. It cannot see which trust stores your visitors or your servers use.</p>
<ol>
<li><span class="ln-state ln-state-detected">Detected</span> <strong>Your site's chain names an emSign or eMudhra issuer.</strong> Get a replacement certificate from another CA's hierarchy and deploy it.</li>
<li><span class="ln-state ln-state-unknown">Unknown</span> <strong>You cannot tell who issued the chain, or you run servers and apps with their own trust lists.</strong> This is the common case. Ask your certificate provider which hierarchy signed your certificate, and check each runtime and image.</li>
<li><span class="ln-state ln-state-confirmed">Confirmed by owner evidence</span> <strong>You hold a certificate issued under a different hierarchy and the package or image versions that carry the trust-store update.</strong> Record the dates. Re-check any stored certificate that chains to emSign ECC Root CA - G3.</li>
</ol>
<p>No scan, ours included, moves you from "unknown" to "confirmed".</p>
</div>

## What should a site owner do this week?

The CERT-In advisory recommends reviewing certificates that chain to the affected CAs, replacing them from unaffected hierarchies, and watching Certificate Transparency logs for unauthorised issuance. When you replace a certificate, check [whether the new one still works for mTLS client authentication](/learn/certificates-losing-clientauth-will-mtls-break). In practice:

- [ ] Find the issuer of every certificate you serve, including mail, API and staging hosts.
- [ ] Replace any that chain to the emSign intermediates named in the Bugzilla record. If eMudhra issued yours, claim the free replacement: it is leaving the global SSL/TLS business and [offers Sectigo certificates through CertiNext](https://www.csoonline.com/article/4239634/emudhra-quits-global-tls-certificate-market-after-compromise-of-ca-signing-keys-2.html).
- [ ] Update browsers, operating system CA packages and runtimes. Rebuild containers from base images that carry the update.
- [ ] Check anything that pins or bundles a CA list, because those do not update on their own.
- [ ] Watch Certificate Transparency for certificates you did not request, and [restrict who may issue with a CAA record](/learn/registry-hijacks-rogue-google-tls-certs-caa-records).

If you buy certificates through a host or an agency, send this:

> For **[domain]**, which CA and which intermediate signed our current certificate? Does it chain to any emSign intermediate or to emSign ECC Root CA - G3? If so, when will it be replaced?

## What does this incident not tell us?

The sources we reviewed do not confirm that anyone used these keys to forge a certificate. CERT-In reports eMudhra's statement that none of the affected intermediates issued end-entity certificates, and Google's report that no TLS certificates chaining to the six TLS intermediates were in Certificate Transparency logs before its block. Both are statements, not proof that nobody forged one, because a recoverable key could sign certificates that never appear in a log you watch.

For more certificate guides, see [the other TLS articles](/learn/topics/tls). For what outside checks cannot prove, read [the limits of an outside scan](/learn/what-an-external-website-scan-cant-prove).

## Questions people ask

### How does a certificate authority generate its private keys?

Inside a hardware security module, during a scripted ceremony that is witnessed or recorded and then reviewed by an auditor. The script is written down beforehand and the output is checked against it. The CA/Browser Forum and Mozilla set these expectations.

### What went wrong with the eMudhra keys?

Eight intermediate CA keys on the P-384 curve had private key values below 256, so they could be recovered almost instantly from the public keys. eMudhra said a flaw in the key generation script produced a second, weaker key pair that was used by mistake. The P-384 curve itself was not broken.

### Is my website affected by the eMudhra incident?

Only if a certificate on your site chains to one of the affected emSign intermediates. Look at the issuer in your certificate chain, or ask your certificate provider. If it is an emSign certificate, replace it from another CA's hierarchy and keep your browser and server trust stores updated.

### Does this mean TLS or elliptic curve cryptography is broken?

No. The curve and the algorithm worked as designed. The keys were not random. A key chosen from a few hundred possible values gives no protection on any curve or key size.

### Do I need to rotate my passwords because of a CA key incident?

Not on the strength of this incident alone. The risk is a forged certificate for a site, used by someone who can intercept a connection. Keep browsers, operating systems and runtimes updated, and investigate unexpected certificate warnings.

### Can a website scan tell me if my certificate chain is affected?

A scan can read the chain your site serves and show its issuer, protocols and expiry. It cannot see what trust stores your visitors' devices or your own servers use. That part needs evidence from the machines involved.


## Sources

1. [Mozilla Bugzilla 2083449: eMudhra intermediate CA incident report](https://bugzilla.mozilla.org/show_bug.cgi?id=2083449): Timeline, the flaw in the key generation script, Mozilla's independent key recovery, the open questions to the CA, and eMudhra's late-October answers. Checked 2 November 2026.
2. [Google Security Blog: protecting Chrome and Android users from compromised eMudhra intermediate CA certificates (20 October 2026)](https://blog.google/security/protecting-chrome-and-android-users-from-compromised-emudhra-intermediate-ca-certificates/): CRLSet block on 17 October, the Android update on 19 October, and the Chrome Root Store removal set for 27 October.
3. [CSO Online: eMudhra quits global TLS certificate market after compromise of CA signing keys (23 October 2026)](https://www.csoonline.com/article/4239634/emudhra-quits-global-tls-certificate-market-after-compromise-of-ca-signing-keys-2.html): Reports eMudhra's 22 October press note and customer update: exit from global SSL/TLS, free replacements, Sectigo certificates through CertiNext.
4. [Mozilla Security Blog: protecting Firefox users following the compromise of eMudhra intermediate certificates (20 October 2026)](https://blog.mozilla.org/security/2026/10/20/protecting-firefox-users-following-the-compromise-of-emudhra-intermediate-certificates/): OneCRL block, Firefox 158.0.1 and the ESR releases, NSS 3.131.1.
5. [CERT-In advisory CICA-2026-3649 (21 October 2026)](https://www.cert-in.org.in/s2cMainServlet?CACODE=CICA-2026-3649&pageid=PUBADV01): Scope, the weak-key description, the vendor's explanation, and the Chrome Root Store date.
6. [Red Hat security bulletin RHSB-2026-012](https://access.redhat.com/security/vulnerabilities/RHSB-2026-012): ca-certificates errata and blocklist guidance for RHEL. Last updated 23 October 2026.
7. [CA/Browser Forum Baseline Requirements for TLS Server Certificates, version 2.3.1](https://cabforum.org/working-groups/server/baseline-requirements/requirements/): Section 6.1.1.1 (CA key pair generation), 6.1.5 (key sizes and curves), 6.1.6 (public key quality).
8. [CA/Browser Forum Network and Certificate System Security Requirements, version 2.0.5](https://cabforum.org/working-groups/netsec/requirements/)
9. [Mozilla Root Store Policy, version 3.1](https://www.mozilla.org/en-US/about/governance/policies/security-group/certs/policy/): Auditor-witnessed key generation ceremony report, annual audits, incident reporting.
10. [NIST SP 800-57 Part 1 Rev. 5: Recommendation for Key Management](https://csrc.nist.gov/pubs/sp/800/57/pt1/r5/final)
11. [NIST SP 800-90A Rev. 1: Random Number Generation Using Deterministic Random Bit Generators](https://csrc.nist.gov/pubs/sp/800/90/a/r1/final)
12. [NIST FIPS 186-5: Digital Signature Standard](https://csrc.nist.gov/pubs/fips/186-5/final)
13. [Debian Security Advisory DSA-1571-1: OpenSSL predictable random number generator](https://lists.debian.org/debian-security-announce/2008/msg00152.html)
14. [The Return of Coppersmith's Attack (ROCA), CCS 2017](https://crocs.fi.muni.cz/public/papers/rsa_ccs17): Weak RSA key generation in a vendor library, a second precedent.
