The safe root stopped being safe
A month ago the advice for anyone caught in the eMudhra mess was simple. The compromised keys sat under emSign ECC Root CA - G3. The RSA root, emSign Root CA - G1, was clean. Mozilla said it wasn't distrusting it, eMudhra moved roughly 30,000 affected certificates onto it, and we told you, in our post on the key compromise and again in the macOS follow-up, to leave G1 alone in your trust stores.
That advice held for about two weeks.
Today, 16 November, G1 gets cut off. It's a mess. Chrome and Firefox stop trusting new G1 certificates tonight at 23:59:59 UTC. Microsoft is pulling the root out of Windows and Edge altogether, starting today, and Apple follows on 30 November if eMudhra's timeline is right. So if you run a certificate that chains to G1, whether it works depends on who's asking.
Four platforms, four kinds of distrust
Here's what each root program announced. The dates line up. The mechanisms don't, and the mechanism decides whether your certificate breaks.
| Platform | What happens | When | Source |
|---|---|---|---|
| Chrome 153+ | Certificates whose earliest SCT is after the cutoff are not trusted by default | SCTs after 16 Nov 2026, 23:59:59 UTC | Chrome Root Program, 4 Nov |
| Firefox (NSS) | TLS distrust-after date set on the root, S/MIME trust removed | 16 Nov 2026, 23:59:59 UTC | Bugzilla 2089115 |
| Windows, Edge, WebView2 | Root removed outright, along with nine other eMudhra roots | Starting 16 Nov 2026 | Microsoft Trusted Root Program, 10 Nov |
| Apple | G1 removed from /etc/ssl/cert.pem in macOS 27.2.1, 26.7.3 and 15.8.3; full removal claimed for 30 Nov | 9 Nov (CLI), 30 Nov (claimed) | Apple, 9 Nov, eMudhra update 8 |
About that last row. Apple's own support note covers the command-line trust store and lists G1 among the removed roots. The 30 November date for Safari and the system trust store comes from eMudhra's incident page and InCommon's update. I couldn't find an Apple document that states it. Plan for it anyway.
Chrome's stated reason: eight eMudhra subordinate CAs generated keys with near-zero entropy, and eMudhra didn't give adequate answers about how it generates keys. Nobody showed that the G1 hierarchy itself was broken. Chrome simply stopped believing the operator, and honestly, fair enough. (HN discussion.)
Why Chrome still says you're fine
Chrome's cutoff keys on the earliest Signed Certificate Timestamp, the proof a CT log hands back when the precertificate gets logged. notBefore doesn't matter here, because a CA can backdate it. Nobody can backdate a log's signature. If the earliest SCT is on or before 16 November 23:59:59 UTC, Chrome keeps trusting the certificate until it expires.
For every G1 certificate that exists today, then, Chrome changes nothing. You can see what Chrome sees:
host=example.com
openssl s_client -connect "$host:443" -servername "$host" </dev/null 2>/dev/null \
| openssl x509 -noout -text \
| grep -E 'Not Before|Timestamp :'
I ran this against a G1 certificate this morning, on India's national scholarship portal. notBefore is 16 September, and all three embedded SCTs are from the same second. Chrome will load that site until the certificate expires on 15 December. Windows won't.
Two caveats. SCTs can also arrive in the TLS handshake or a stapled OCSP response, and this command won't show those, though most public certificates embed them. And Chrome lets an enterprise override the constraint: if the root is explicitly trusted on the platform, say through a Windows Group Policy Object, the SCT check is skipped. Don't build a plan on that.
Windows breaks first
Microsoft went blunt. Its Trusted Root Program announcement lists ten roots, including emSign Root CA - G1 (SHA-1 thumbprint 8AC7AD8F73AC4EC1B5754DA540F4FCCF7CB58E8C) and emSign Root TLS CA - G1. Windows removes them through the automatic root update endpoint. Edge and WebView2 lose default trust through a PKI Metadata component update. Microsoft's guidance: replace any certificate chaining to these roots before the change reaches your devices.
No SCT grace period. Once the update lands, a G1 chain fails in Edge and in WebView2 apps. It also fails in anything that validates through Windows' own certificate APIs, which covers curl.exe on Windows (it uses Schannel) and plenty of .NET and PowerShell code.
The annoying part is propagation. Root updates reach machines at different speeds, so for a while some of your users get an error and some don't. Your support queue will know before your dashboards do.
Here's the chain that scholarship portal serves, trimmed to the common names:
0 s:CN = scholarships.gov.in
i:CN = EM OV TLS CA - G2A-1
1 s:CN = EM OV TLS CA - G2A-1
i:CN = emSign Root TLS CA - G1
2 s:CN = emSign Root TLS CA - G1
i:CN = emSign Root CA - G1
Both of Microsoft's G1 roots in one chain. Grep for one name and you miss half of it.
Your monitoring can't express any of this
This is the bit that bothers me. A distrust-after date is metadata on a root, and a PEM bundle has nowhere to put it. A file like /etc/ssl/certs/ca-certificates.crt either contains G1 or it doesn't.
Look at curl's Mozilla extract. The current cacert.pem, built from Mozilla data as of 5 November, still has emSign Root CA - G1 as a plain trusted root. I grepped mk-ca-bundle.pl, the script that builds it. It knows about "Explicitly Distrust" blocks and ignores distrust-after completely. certifi put G1 back on 19 October after briefly dropping it. Keeping the root is right for compatibility and wrong for a cutoff, and a PEM file gives you no third option.
So your OpenSSL-based uptime check keeps calling the G1 certificate valid. So does the Python script running on requests, and the agent on your Linux box. Meanwhile your Windows users are staring at an error page, and your Mac users join them on 30 November. We saw the mirror image of this with Chrome's CRLSet and GoDaddy's revocations. Back then the browser was the lenient one. This time it's your monitoring.
"Valid chain" only means something relative to a trust store. Which one does your monitoring use? If you don't know, you have the same blind spot as checking from a single region, only across trust stores this time.
Find every G1 certificate you own
Skip the inventory. Nobody's inventory is complete, and G1 certificates went out as an emergency fix, which is exactly when nobody updates the spreadsheet.
Start with CT. Every one of these certificates was logged, so crt.sh knows about them even if you don't:
domain=example.com
curl -s "https://crt.sh/?q=%25.${domain}&output=json&exclude=expired" \
| jq -r '.[] | select(.issuer_name | test("eMudhra"))
| [.not_before, .not_after, .common_name, .issuer_name] | @tsv' \
| sort -u
That catches anything eMudhra issued for your names, G3 leftovers included. Then check what your endpoints actually serve. A certificate that got replaced in CT can still be sitting on some forgotten box:
#!/usr/bin/env bash
# Flag hosts whose served chain runs up to eMudhra's emSign G1 roots.
# usage: ./g1-check.sh hosts.txt (one hostname per line)
while read -r host; do
[ -z "$host" ] && continue
chain=$(timeout 10 openssl s_client -connect "$host:443" -servername "$host" \
-showcerts </dev/null 2>/dev/null)
if [ -z "$chain" ]; then
echo "$host NO-TLS-RESPONSE"
elif grep -qE 'emSign Root (TLS )?CA - G1' <<<"$chain"; then
expiry=$(openssl x509 -noout -enddate <<<"$chain" | cut -d= -f2)
echo "$host EMSIGN-G1 expires: $expiry"
else
echo "$host ok"
fi
done < "$1"
Point it at every IP behind your load balancers as well as the hostname. One stale node is enough.
Want to know what Windows sees? Test from Windows. Run curl.exe -sS -o NUL https://your-host/ on a machine that has picked up the root update and a G1 chain should fail with a Schannel error. A Linux box will never tell you that.
Universities are moving for the third time this year
For the version at scale, look at InCommon, the certificate service a lot of US universities buy through. Its eMudhra-operated issuing CA ran on G1. InCommon now calls it "no longer viable for public trust", started issuing DV certificates through SSL.com on 9 November, and lists 16 November as the day eMudhra stops issuing on G1.
Todd Gardner at CertKit counted the damage in the CT logs on 12 November. 59 large universities held 36,824 certificates on G1. 21,462 of them, 58%, hadn't been replaced by a certificate from another CA yet. That was four days before Windows pulled the root. For those schools it's the third forced move of 2026: off Sectigo onto eMudhra, off G3 onto G1, and now onto SSL.com.
That's what you get when every migration means grabbing whatever is trusted this week. Nobody ever made switching CAs boring.
Make the next CA switch boring
You'll do this again. Maybe not with eMudhra, but some CA you depend on will have a bad month while you're a customer. There's no perfect CA to pick. What you can do is turn a CA change into a config change.
Issue over ACME wherever you can, so moving CA means a new directory URL and an account, with no procurement cycle. eMudhra said about 97% of the affected certificates were issued over ACME, and that's the only reason a move like this is survivable at all. Keep a second CA's ACME account registered and actually tested; we made that case in the post on single-CA risk. Put both CAs in your CAA records before you need them. If you use CAA account binding, bind both accounts, because a CAA record that only names the CA you're fleeing will block you on the day you have to leave. And monitor against Mozilla's, Windows' and Apple's trust stores. The one your monitoring box happened to ship with is a third of the picture.
As for G1: replace those certificates now. "Chrome trusts it until December" is true and almost useless. Edge stopped trusting it today.
Frequently asked questions
Does Chrome stop trusting my existing emSign G1 certificate on 16 November? No. Chrome only distrusts G1 certificates whose earliest SCT is dated after 16 November 2026, 23:59:59 UTC, so a certificate logged before then keeps working in Chrome until it expires. Windows and Edge are a different story, because Microsoft removes the root itself.
Why does Microsoft break existing certificates when Chrome and Firefox don't? Microsoft removed the ten remaining eMudhra roots from its trust list and set no cutoff date, so certificates issued before 16 November get no grace period. Its guidance is to replace affected certificates before the update reaches your devices.
Is my server affected if it only sends the leaf and the intermediate? Yes, if the chain ends at G1. Clients build the rest of the chain from their own trust store, so what counts is the root the chain ends at, whatever you send. If your intermediate's issuer is emSign Root CA - G1 or emSign Root TLS CA - G1, you're on the hierarchy being removed.
Will curl and Python stop trusting G1 certificates after the cutoff? Not on their own. PEM bundles like curl's cacert.pem and certifi can't express a distrust-after date, so they either keep the whole root or drop it. The 5 November curl extract still has G1 as a fully trusted root, so non-browser clients on Linux keep accepting G1 chains until a bundle update removes it.
Can I still get a new certificate from eMudhra on G1? Don't try. eMudhra and InCommon both list 16 November as the end of G1 issuance, and a certificate logged after the cutoff is distrusted by Chrome and Firefox and fails on Windows no matter when it was issued.