Domino's eMudhra Certificate Alert: What the 9.1 Really Means, and Which Trust Store to Clean
You run a few Domino servers. In late October an HCL security bulletin names Domino, and next to the CVE sits a red 9.1 / 10. Your boss forwards the link and asks whether this is an all-nighter.
Breathe first. This is not the kind of hole where someone runs code on your Domino box from across the internet — Domino’s own code is fine. What broke is a set of widely trusted CA certificates: a certificate authority called eMudhra has several intermediate CAs whose private keys can be recovered, and those CAs are in turn trusted by Java’s trust store and by the chain Domino uses for outbound connections. In other words, this is a wrong-thing-in-the-trust-store problem, not an execution-layer break. The score looks frightening; the work is concrete and very much under your control.
This piece covers three things: how to read that 9.1, whether your environment is really exposed, and which certificates to pull from which trust stores.
TL;DR
- The nature of it: the private keys of eMudhra intermediate CAs can be recovered. Anyone holding them can issue a “looks legitimate” TLS certificate for any domain. As long as an attacker can sit in the middle of your connection (a man-in-the-middle, MITM), a Java TLS client that still trusts those CAs will accept the forged site.
- Not a bug in Domino’s code — a problem with what’s inside the trust store. The CWE sits under Improper Certificate Validation1.
- The 9.1 has two readings: Oracle and HCL both score it CVSS2 9.1; Red Hat scores it 6.8 on CVSS 3.1 — the difference is that the real MITM path requires the attacker to first hold a middle position and the victim end to connect out. Neither reading is “someone is already on your server.”
- Are you exposed? Look in the right place: on most Domino servers the CAs are trusted in the bundled Java trust store (
cacerts) and in the oldercacert.pem;certstore.nsfholds only the Let’s Encrypt roots by default, and a standard Domino Directory doesn’t carry these public roots either — unless you imported them yourself. - What to do: remove the eMudhra / emSign / Comodo roots HCL lists from every trust store, match by fingerprint, not by name, and remove all of the related ones (pull just one and the chain can rebuild through another path).
- Affected versions (HCL’s Domino Server list): 12.0.2 FP8 and earlier, 14.0 FP5 and earlier, 14.5 up to 14.5.1 FP1. Both R12 and R14 are in scope.
What the vulnerability actually is
TLS trust is a chain: when your server connects out, the far end presents a certificate, that certificate is signed a layer up, and so on, until it reaches a root CA in your trust store. If the chain reaches a root you trust, the connection is treated as safe.
What broke this time is the middle of the chain. The heart of CVE-2026-91403 is that root CAs in Java’s trust store signed eMudhra intermediate CAs whose private keys can be recovered. Community analysis points to 8 affected eMudhra intermediates (6 signing TLS, 2 signing code), all cross-signed under the AAA Certificate Services root.
Holding an intermediate CA’s private key is a master key: you can issue a certificate for mail.yourcompany.com, login.partner.com, any domain, that chains back to a trusted root. Then an attacker who can sit in the middle of the connection (reroute, poison DNS, ARP on the same segment…) feeds the forged site to your Domino, and a TLS client that still trusts those CAs accepts it.
Worth stressing: revocation alone won’t stop this. Many TLS implementations don’t check revocation status by default, so the fix is to remove these CAs from the trust store outright, not to hope a revocation list catches them.
9.1 or 6.8?
You’ll see two scores for the same CVE. Both are right; they measure different things:
| Scorer | Score | Vector | Measuring what |
|---|---|---|---|
| Oracle / HCL | 9.1 | AV:N/AC:L/PR:N/UI:N/... (high C, high I) | the broad JSSE / untrusted-code worst case |
| Red Hat | 6.8 | AV:N/AC:H/PR:N/UI:R/... | the real MITM-TLS path: needs a middle position + interaction |
The gap is in AC (attack complexity) and UI (whether user/client interaction is needed). Oracle scores against the widest framing; Red Hat scores the practical “forge a cert → man-in-the-middle” path, which lands medium-high rather than critical. For a Domino admin that means: fix it, but it’s not “someone is hammering your 443 with an exploit right now.” High priority — with time to do it correctly.
Is your environment actually exposed
This is where the 9.1 tempts people into overreacting. The question isn’t “is Domino affected,” it’s “which trust store actually holds these CAs.” Domino touches several, and their defaults differ a lot:
- The bundled Java trust store (
cacerts) — the main place it burns. Domino’s bundled JVM ships its own trust store (PKCS12 on 14.x, still namedcacertwith no extension) that includes these public roots. This one almost always needs attention. cacert.pem/cacerts.pem(data directory) — the older trust source. From Domino 14.5, server-sideNotesHTTPRequestreads roots from the Domino Directory by default and no longer reads this file — but if your environment still relies on it (or uses the.iniswitch back to the old behaviour), clean it too.certstore.nsf— community testing (Daniel Nashed, late October) notes it holds only the Let’s Encrypt roots by default, so it’s in scope only if you imported these public CAs yourself.- A standard Domino Directory (
names.nsf) — also doesn’t carry these public roots by default; again, only relevant if you added them by hand.
The practical takeaway: check the bundled Java cacerts first, then look at whether you’ve touched cacert.pem, certstore.nsf, or the Domino Directory. On a server with an untouched trust source, the thing to clean is usually that bundled Java store. This is also why anyone who already understands what each of the trust stores is for has an easy time here.
Don’t assume about EOL versions: one measured 9.0.1
The instinct is “9.x is older, so it must carry the same certs as 12/14” — but the measurement showed otherwise: 9.x is affected too, just by AAA rather than emSign. We checked all three trust stores on a 9.0.1 FP10 Domino server — a Domino 9 container built from our public build recipe (the site has a piece on how to build it); IBM J9 Java 1.8.0_151, cacerts dated 2018, 105 entries, which you can reproduce from the same recipe:
- The bundled JVM
cacerts(checked withkeytool -list— the authoritative one): no emSign / eMudhra at all; but Comodo’s AAA Certificate Services is present (aliascomodoaaaca, SHA-256D7:A7:A0:FB…69:EF:4). names.nsf(Domino Directory): the grep turned up the Comodo AAA string — consistent with HCL, which says Comodo AAA is pre-populated in names.nsf by default. Per HCL, that’s the cert you remove from names.nsf (Security→Certificates).cacert.pem: this box doesn’t have the file at all, so the tools that targetcacert.pemhave nothing to act on here.
Why no emSign in any of the three? This JVM’s cacerts is old (2018), and emSign was only added to trust stores around 2018 onward — so it was never bundled in, consistent with the official lists stopping at 12.0.2 / 14.x for emSign.
Lining this up with HCL’s steps matters: HCL’s Java keystore step removes emSign G3 — 9.x has no emSign in the keystore, so the Comodo AAA that shows up in the bundled JVM is not on HCL’s keystore-removal list; the Comodo AAA HCL removes lives in names.nsf (pre-populated) and cacert.pem. So a 9.x server done HCL’s way = leave the bundled JVM alone, delete the Comodo AAA from names.nsf (and cacert.pem if present).
Scope note: this is one measured 9.0.1 FP10 server container; a 9.0.1 whose JVM was manually updated could differ. The one solid takeaway: don’t skip 9.x, but don’t copy the 12/14 list onto it either — run
keytool -listby fingerprint on each of your own 9.x boxes and sort out which store holds what.
Which certificates to remove
HCL’s manual-remediation guidance lists several roots to pull. Below, as of writing (2026-11-05), is the list and the SHA-256 fingerprints as relayed by the community (Nashed, panagenda) — but always go by the current list in HCL’s bulletin, since HCL may update it as the investigation moves:
emSign ECC Root CA - G3 (the main one) SHA-256 86:A1:EC:BA:08:9C:4A:8D:3B:BE:27:34:C6:12:BA:34: 1D:81:3E:04:3C:F9:E8:A8:62:CD:5C:57:A3:6B:BE:6B
emSign Root CA - C1 SHA-256 12:56:09:AA:30:1D:A0:A2:49:B9:7A:82:39:CB:6A:34: 21:6F:44:DC:AC:9F:39:54:B1:42:92:F2:E8:C8:60:8F
emSign ECC Root CA - C3 SHA-256 BC:4D:80:9B:15:18:9D:78:DB:3E:1D:8C:F4:F9:72:6A: 79:5D:A1:64:3C:A5:F1:35:8E:1D:DB:0E:DC:0D:7E:B3
AAA Certificate Services (Comodo) SHA-256 D7:A7:A0:FB:5D:7E:27:31:D7:71:E9:48:4E:BC:DE:F7: 1D:5F:0C:3E:0A:29:48:78:2B:C8:3E:E0:EA:69:9E:F4Two operational points that people who’ve done it will press on:
- Match by fingerprint, not by name. The same root shows up under different display names across trust stores (the AAA root may read as “Comodo AAA Services root” in
cacert.pem). Comparing by SHA-256 fingerprint is what keeps you from missing one or deleting the wrong thing. - Remove all of them, not some. Nashed calls this out specifically:
emSign ECC Root CA - G3and the Comodo cross-cert both have to go — pull only one and the chain can rebuild through the other path, which is the same as not cleaning at all.
A contested fifth: emSign Root CA – G1. The four above are the HCL/Domino community (panagenda) removal list; G1 isn’t on it — as of 2026-11-05, Mozilla chooses not to remove it and Chrome is still reviewing (G1’s issue leans on a CRL-signature problem, not quite the same line as the eight intermediate-CA key compromises). But some Java-trust-store guidance (Ignition, Camunda) and some field fix tools take the more conservative stance and remove G1 too. If you hit G1 while cleaning a Java
cacerts, it’s not an omission — it’s vendors drawing the line differently; go by your vendor’s current guidance, and test the TLS chains your services depend on before removing it.
How to remediate on the server: steps and syntax
The authoritative steps are HCL KB0134844 (“What steps are required to protect HCL Notes and Domino from the eMudhra Certificate Compromise issue”) — as of 2026-11-05 it’s manual steps only, no Fix Pack. Order: check → back up → remove per store → verify. HCL is explicit about which store holds which cert.
1. Check first (read-only, changes nothing) — use the bundled keytool to list the JVM cacerts and match fingerprints:
"<Domino>\jvm\bin\keytool" -list -v -keystore "<Domino>\jvm\lib\security\cacerts" -storepass changeit | findstr /i "emSign emudhra 86:A1:EC BC:4D:80 12:56:09 D7:A7:A0"On Linux/AIX replace findstr with grep -i; on Domino 14.5+ use keytool -list -v -cacerts (no path or password).
2. Back up (a corrupted cacerts means this box’s HTTPS/TLS won’t start):
copy "<Domino>\jvm\lib\security\cacerts" "<Domino>\jvm\lib\security\cacerts.bak"3. Remove per store, the way HCL maps it:
- Java keystore
cacerts→ remove emSign ECC Root CA - G3 (bundled keytool; the alias differs by version — 12.0.2 uses the full DN with-keystore cacerts -storepass changeit, 14.x uses the-cacertsshortcut):
"<Domino>\jvm\bin\keytool" -delete -alias "cn=emsign_ecc_root_ca_-_g3,o=emudhra_technologies_limited,ou=emsign_pki,c=in" -keystore cacerts -storepass changeitcacert.pem(data directory; libcurl uses it for LotusScript / OIDC / DominoIQ) → delete the three emSign PEM blocks (G3 / C1 / C3); on older versions also delete the Comodo AAA block.names.nsf(Domino Directory) → delete the Comodo AAA cert. HCL says it’s pre-populated in names.nsf by default: open names.nsf → Security→Certificates → selectCN=AAA Certificate Services/O=Comodo CA Limited/…and delete it → replicate across the domain and restart one server at a time.certstore.nsf/ anything else: only if you manually added other affected certifiers.
⚠️ Comodo AAA is a broadly-used root with a large regression surface — before removing it from cacert.pem / names.nsf, test the TLS chains you depend on (next section); restart Domino / the HTTP task after changes so they reload.
4. Verify: re-run step 1 to confirm the keystore is clean; the other stores and the live-connection check are in the next section.
How to confirm it’s actually clean
The most direct check, and one every version has: re-run the keytool -list from step 1 and confirm the cert is gone from cacerts — you changed this store, so checking it directly is the surest test. 9.x doesn’t have the LotusScript test below, so this is the one to use there.
To go further and test an actual connection, use emudhra.we-have-a-problem.com, maintained by Mozilla engineer Dennis Jackson and the community (open source, Apache 2.0, with a checker, per-OS workarounds, and a list of patched software). That page is the checker/docs; the actual targets are the per-intermediate test hosts it lists, e.g. dv-g3b-emudhra.we-have-a-problem.com and dv-g3b-aaa-emudhra.we-have-a-problem.com — each should fail with a certificate error; if it connects and loads, that runtime still trusts the bad CA. Test at least one -emudhra and one -aaa-emudhra (a store can distrust one path while still trusting the other — exactly the emSign-vs-AAA split). On V10+ you can also hit these hosts from the server with Nashed’s LotusScript snippet:
Dim req As New NotesHTTPRequestreq.Maxredirects = 0Dim response As Stringresponse = req.Get(testUrl) ' once clean, this should FAIL (cert not trusted)Two things to keep straight: whatever makes the connection is the trust store you’re testing. NotesHTTPRequest only exists on V10+ (9.x doesn’t have it) — on 9.x you’d test an actual connection with a Java test running under Domino’s JVM instead; and curl tests the OS’s (or its own) CA store, not Domino’s cacerts — a curl pass doesn’t mean Domino is clean. For Domino’s JVM store go by keytool -list (for cacert.pem you can point curl at that file with curl --cacert cacert.pem <url>). Every component that connects out uses its own trust store, so confirm them separately.
How this string of CVEs maps together
You’ll run into more than one CVE number here, which is easy to tangle. Pulled apart, it’s clear:
- CVE-2026-91403 / CVE-2026-114820 — the upstream Java problem (Oracle Java SE, GraalVM, OpenJDK). 91403 was the first fix (distrust only the intermediates); because it turned out easy to bypass and the first fix wasn’t enough, OpenJDK removed the root CAs outright, which got the new number 114820. Domino bundles the JVM, so it’s affected along with it — which is why the Domino Server bulletin (KB0134847) carries both numbers.
- CVE-2026-108014 — HCL’s own number, for the certificate trust in Domino Server’s outbound-request features; the cve.org record shows it published 2026-10-27 with a 9.1 CVSS. This is the one behind that “official bulletin” at the end of October.
- CVE-2026-108015 — same nature, but for the Notes client’s outbound features.
Four numbers, two layers (upstream Java + Domino itself) — but the closing move is the same one thing: get those affected roots out of your trust stores.
One line to close
This is a textbook “high score, but the substance is trust-store hygiene” event. It follows the same thread as Domino 14.5 moving NotesHTTPRequest’s trust source to the Domino Directory, and the certstore / CertMgr certificate model — the same world where DRAPI reads only the JVM truststore. Whoever actually knows which roots their server trusts, and in which stores, is just cleaning a few certificates by fingerprint this week; whoever doesn’t is the one that 9.1 leaves frozen, unsure where to start.
Footnotes
-
CWE (Common Weakness Enumeration) classifies the type of weakness. This one sits under CWE-295 (Improper Certificate Validation) and CWE-327 (Use of a Broken or Risky Cryptographic Algorithm) — both pointing at a broken trust relationship rather than an execution-layer break like a memory compromise. ↩
-
CVSS (Common Vulnerability Scoring System) is the industry-standard severity score for vulnerabilities, 0–10, with a “vector string” spelling out attack path, complexity, privileges required, and so on. Different scorers can land on different numbers for the same flaw because they assume different attack scenarios. ↩
Sources
- Security Bulletin: eMudhra Compromised Certificate vulnerabilities in Java impact HCL Domino Server (CVE-2026-91403, CVE-2026-114820) — HCL Support (official)
- CVE-2026-91403 — official CVE record (cve.org, Oracle CNA)
- CVE-2026-108014 — official CVE record (cve.org, HCL Domino Server)
- CVE-2026-91403 — Red Hat Customer Portal (the CVSS 6.8 reading and how the fix evolved)
- eMudhra CA Key Compromise — How to remove the affected certs — Daniel Nashed's Blog (Domino community)