Removing eMudhra from a browser or Windows trust list does not remove it from every system that validates certificates. Organisations should treat the response as controlled trust-store remediation: map the validators behind critical services, replace affected certificate chains and test changes before withdrawing trust.
Cisco’s advisory, materially updated on 20 November, remains an operational reference for Cisco estates. Its product matrix is still moving: some cloud services are remediated, some appliances need a specific update or trust-store action, and a substantial group remains under investigation. There is no single version number that fixes a certificate authority (CA) compromise. What matters is the certificate bundle and validation path each system actually uses.
For UK and Channel Islands organisations, this is as much an asset-management and change-control incident as a security-update incident. Public-facing services may need replacement certificates where they chain to withdrawn hierarchies. Less visible systems need checking for their own CA copies: Java cacerts files, container base images, VPN and firewall appliances, integration gateways, developer tooling, embedded devices and software that ships a private trust bundle.
Known exploitation status: Cisco said on 20 November that it was not aware of malicious use of the issue. That is reassuring, but it does not lower the priority. A fraudulent Transport Layer Security (TLS) certificate is most useful to an attacker who can also influence or intercept a connection; a fraudulent code-signing certificate may support a narrower, harder-to-observe software supply-chain attack. Neither makes the underlying trust failure routine.
What has changed since the first emergency response
The confirmed core incident remains grave but specific: private keys for eight eMudhra intermediates were recoverable because of a defective April 2025 key-generation ceremony. Six were TLS issuing CAs for domain-validated, organisation-validated and extended-validation certificates; two were code-signing issuers. Cisco says the affected certificates were cross-signed by Sectigo’s AAA Certificate Services root, extending their reach to systems that did not directly trust an eMudhra root.
The consequential follow-on action is at root-store level. Microsoft first blocked the affected authorities through its disallowed-certificate list, then removed six eMudhra G3 roots and the AAA Certificate Services cross-sign. On 16 November, it began removing the remaining ten eMudhra roots from the Microsoft Trusted Root Program and Windows trust list. Edge and WebView2 also removed default trust for remaining roots including emSign Root CA – G1, emSign Root CA – C1, emSign ECC Root CA and emSign Root TLS CA – G1.
That goes further than revoking compromised intermediates. Revocation asks clients to recognise that a particular certificate is no longer valid. Root removal says certificates chaining to that hierarchy should no longer be trusted by default. The difference matters because revocation checking is uneven across real software estates, particularly where devices are offline, validation libraries are old, or applications carry their own certificate bundle.
Mozilla used OneCRL, its certificate-revocation block list, to block the affected certificates in Firefox and released Firefox 158.0.1, Firefox ESR 115.43.1 and ESR 153.5.1 for systems that might not receive the normal block-list update. It also updated NSS 3.131.1, the Network Security Services trust store used by Firefox and some other software. Apple began blocking the affected TLS intermediates automatically on 19 October and later blocked emSign ECC Root CA – G3. It has warned that some command-line tools using macOS’s separate /etc/ssl/cert.pem bundle could remain an exception until a software update reaches them.
There is a second development that is easy to muddle with the first. eMudhra’s later material in Mozilla’s public incident record describes eight additional RSA G2B/G2C subordinate CAs withdrawn on 7 November because eMudhra could not demonstrate compliant handling of the private-key material during the same ceremony. That is a serious assurance failure, but not the same claim as demonstrated recoverability of the original eight ECDSA keys. Treating all 16 as proven brute-force recoverable would overstate what has been established.
How the exposure persists after browsers and Windows update
A browser is only one certificate-validation client, and in many estates not the most consequential one. A workload may validate an upstream API certificate through Java’s bundled trust store; a container may inherit an old CA bundle from its base image; an appliance may use a vendor-curated store updated only with firmware; and a line-of-business application may pin or import trust anchors independently of the operating system.
So “Windows has removed the root” needs qualification. It may protect software using the current Windows certificate APIs and trust list, provided the device receives the change. It does not automatically alter every application-managed store on that PC, every Windows Subsystem for Linux (WSL) instance, every container image in a registry, or every Linux VM and appliance elsewhere in the organisation.
Microsoft has documented that unevenness. Azure Linux images were refreshed from 16 November, while .NET container images depend on upstream certificate-package updates and the .NET SDK trust bundle is scheduled separately. The Microsoft Build of OpenJDK follows the Mozilla trust-store update cycle. Assess Node.js, Java, Python distributions, CI runners and vendor appliances against their own release notes and shipped trust bundle, not the host operating system.
There is an availability trade-off. Removing eMudhra trust can break legitimate services still presenting a certificate that chains to a withdrawn hierarchy. That is not a reason to restore broad trust. Identify the dependency, replace the certificate or chain with one from an unaffected provider, test it with the client population that matters, and make any narrowly scoped temporary exception visible to the security owner and expiry-controlled.
Editorial view: This incident exposes the gap between patch management and trust management. A monthly patch dashboard can look healthy while a forgotten Java process, long-lived Kubernetes image or branch-office appliance continues trusting a hierarchy that mainstream platforms have abandoned. Do not panic-remove certificates from every system. Find where validation occurs, then remove trust and replace dependencies with evidence.
What good looks like now
The difference matters because revocation checking is uneven across real software estates, particularly where devices are offline, validation libraries are old, or applications carry their own certificate bundle.
Start with a certificate dependency inventory, not a search for “eMudhra” in endpoint logs. For internet-facing services, inspect deployed certificate chains and certificate-management records. For client systems, identify the stores actively used by browsers, runtime environments, application servers, containers, developer tools and network or security devices.
Certificate Transparency logs are useful: they provide a public record of certificates issued for internet names and can identify unexpected issuance for domains an organisation controls. They cannot show private PKI certificates, internal device trust stores or every code-signing decision. They are a detection source, not an inventory substitute. The NCSC’s guidance stresses both issuance monitoring and the limits of revocation in the Web PKI, the public-key infrastructure used for web certificates.
For code signing, review validation controls in endpoint protection, application control and build or release pipelines. A signature rooted in a generally trusted chain must not override provenance checks, publisher allow-lists, repository controls or behavioural detection. The compromise creates a route by which malicious software could look more legitimate; it does not make every signed binary suspicious by default.
Priority actions for the next few days
- Confirm that browser and operating-system trust changes have landed. Pay particular attention to disconnected Windows devices, gold images, Azure scale sets, virtual desktop infrastructure (VDI) pools and machines with automatic root updates disabled.
- Check public TLS chains and renew affected certificates early. Do not wait for expiry if a live service chains to an eMudhra hierarchy being removed. Test from the client types that matter, including mobile apps, integrations and older appliances.
- Audit separate trust stores. Prioritise Java and OpenJDK runtimes, Linux and container CA packages, Kubernetes workloads, API gateways, proxies, firewalls, VPNs, mail gateways, build agents and embedded or operational technology devices.
- Rebuild, do not merely patch, long-lived images where practical. An updated host does not repair an old CA bundle baked into a container image or virtual appliance template.
- Monitor Certificate Transparency for your domains. Escalate unexpected certificates quickly, preserving the chain, timestamps and issuer information for the CA, registrar, hosting provider and incident-response team.
- Review code-signing exposure proportionately. Check whether security tooling accepts signatures based solely on platform trust, and ensure anomalous publisher certificates or newly observed signing chains receive review.
- Do not create an indefinite local trust exception. If a business-critical service breaks, document the owner, precise certificate, affected clients, compensating controls and an expiry date. Broadly re-importing a removed root defeats the mitigation.
Cisco estates: status is mixed, so use the live matrix
Cisco’s 20 November advisory should be treated as the current product-status page, not a one-off bulletin. It says Secure Email Gateway and Secure Email and Web Manager move to Trusted Root 2.8 automatically where updates are automatic; manually managed deployments require the updatenow action. Webex Calling cloud remediation is marked complete, while Webex Meetings cloud remediation remains in progress with no customer action specified.
Some product fixes are already identified. On-premises RoomOS releases 11.41.2.3 and 26.7.4.3 are listed, as is RoomOS Cloud 26.10.1 for the current generation. Cisco also lists fixed releases or guidance for IOS XE, IOS XR, ISE, Secure Firewall, FMC, NSO, ThousandEyes and many other products. Several target releases are dated after 22 November, including IOS XR 26.4.1 on 30 November and ISE 3.5P5 in December. Other products remain under investigation. “Cisco patched” is not a meaningful estate-wide status.
Work through the current affected-product table, identify every deployed product and version, then apply the specified release, remediation script or certificate-removal guidance. Where Cisco says a cloud service needs no customer action, record that vendor status but still assess integrations and local clients that validate its endpoints. The earlier Cisco emergency distrust guidance still applies, but the practical task has grown from immediate blocking to tracking trust-store convergence across the estate.
Residual risk: what remains unresolved
The platform response has reduced the chance that newly minted fraudulent certificates will be accepted by routinely updated mainstream clients. It does not establish that every enterprise trust store has converged, nor does it retrospectively prove that no fraudulent certificate or signed file was ever used. Cisco’s current position is that it knows of no malicious exploitation, not that exploitation can be ruled out.
eMudhra says it is winding down publicly trusted TLS issuance and has commissioned an independent review. Its stated migration dates are not a universal safety boundary: browser programmes, operating systems and application ecosystems apply distrust through different mechanisms and schedules. Existing certificates, especially those issued before a programme’s cut-off, may receive different treatment from new issuance. Test the actual chain against the actual clients.
The emergency phase is not over for organisations with complex estates. The operational priority is clear: verify trust removal in managed platforms, replace affected service certificates, find independent trust stores, and maintain a short-lived exception process for unavoidable breakage. It is less dramatic than saying “the web is broken”, but it is the work that prevents a CA compromise becoming a quiet, long-tail enterprise exposure.
For organisations still building a certificate inventory, the enterprise clean-up implications of certificate distrust show why certificate ownership, renewal automation and trust-store governance are operational controls, not PKI housekeeping.
Sources and further reading
- Cisco: Informational for Emergency Distrust of eMudhra Root and Intermediate Certificate Authorities
- NCSC: Provisioning and managing certificates in the Web PKI
- Microsoft Security Response Center: Microsoft’s response to compromised eMudhra certificate authority keys
- Mozilla Security Blog: Protecting Firefox users following the compromise of eMudhra intermediate certificates
- Apple Support: About the compromised eMudhra intermediate CAs
- eMudhra: Statement on Recent Certificate Security Incident
- Mozilla Bugzilla: eMudhra private key compromise of intermediate certificates
Spot an error?
If something factual looks wrong, outdated or misleading, flag it here. Corrections are reviewed separately from normal article comments and reader questions.