Skip to content
S1DCreh2Ge

S1DCreh2Ge

hackmd.io • October 28, 2026

A default Teku node on mainnet is not affected and needs no action. Default means no ` URLs in its configuration. Teku is affected only when it connects to an HTTPS endpoint, as listed below.

On 2026-10-17, the private keys of eight eMudhra (emSign) intermediate certificate authorities were shown to be recoverable from their public certificates. They were revoked on 2026-10-18, and browsers, operating systems and Go have since blocked them.

Java does not check revocation by default, so Java runtimes need their own fix. Anyone holding one of these keys can create a certificate that Java accepts as genuine. This includes the Eclipse Temurin runtime in the Teku Docker images.

OpenJDK shipped two fixes: on 2026-10-20 it blocked the compromised intermediates and their Sectigo cross-signs (CVE-2026-91403), and on 2026-10-23 it removed four eMudhra and Sectigo roots (CVE-2026-114820).

On a runtime with neither fix, an attacker who can intercept Teku's network traffic can impersonate an HTTPS endpoint that Teku connects to. Teku does not accept client certificates on any of its APIs, so there is no inbound exposure.

You are affected if Teku makes HTTPS connections to another host. That happens when any of these options holds an ` URL, or when you run one of the listed testnets:

- External signer: `--validators-external-signer-url`, signer URLs added through the key manager API, or `--validators-external-signer-public-keys` set to a URL

- Validator client to beacon node: `--beacon-node-api-endpoint`, or endpoints in `--sentry-config-file`

- Proposer config loaded from a URL: `--validators-proposer-config`

- Builder or relay: `--builder-endpoint` pointing straight at a relay or a remote mev-boost

- Execution client: `--ee-endpoint` on a remote host

- Startup data from a URL: `--checkpoint-sync-url`, `--initial-state`, `--genesis-state`, `--ws-checkpoint`, `--network`, `--p2p-discovery-bootnodes-url`, `--p2p-static-peers-url`

- Testnets: Sepolia, Hoodi and Ephemery download their starting state over HTTPS when Teku starts with an empty database

You are not affected by this issue if all of these are unset, local files, or ` URLs. Peer-to-peer networking (libp2p and discv5) does not use certificate authorities and is not affected. Plain ` to another host has no authentication at all, so prefer HTTPS with a fixed Java runtime.

**Docker users:** every `consensys/teku` image up to and including 26.10.2, including the `-jdk25` and `-jdk26` tags, ships a Java runtime with neither fix: it still trusts the emSign ECC Root CA - G3. Updating Java on the host does not change the Java inside the image. If you run Teku in Docker and use any of the options above, you are affected until a fixed image is released.

### How your environment changes the risk

- **Network path.** Intercepting traffic needs a position on the path, such as shared hosting, a cloud provider, a compromised router or a hostile Wi-Fi network. A node whose endpoints are on localhost or a private network is much less exposed.

- **How Java was installed.** Your Java version alone does not tell you whether you are exposed; the OpenJDK guidance under How to fix your Java runtime explains how to check.

- Docker images: the default `consensys/teku` tags, such as `latest` and `26.10.2`, ship Eclipse Temurin 26, and the `-jdk25` tags ship Temurin 25. Both trust the emSign ECC Root CA - G3 but not AAA Certificate Services.

- Oracle and OpenJDK builds before the fixes include both roots.

- Ubuntu and Debian packages: Java uses `/etc/ssl/certs/java/cacerts`, which takes extra steps to update. See the OpenJDK guidance under How to fix your Java runtime.

- RHEL, Fedora and others: Java uses the operating system's certificate store.

The attacker must be able to intercept traffic between Teku and the endpoint. Anyone can derive the compromised keys from the public certificates, so that network position is the only real barrier. No exploitation has been reported.

| Endpoint | Worst case if impersonated | How likely |

| External signer | Get signatures for voluntary exits (irreversible) and builder registrations that redirect fees. Get slashable signatures if the signer's own slashing protection is off. Capture credentials embedded in the signer URL. | Low to medium. Most signers run on localhost or a private network; a signer reached over the internet is more exposed. Mutual TLS prevents it. |

| Beacon node (for a validator client) | Have validators propose blocks that pay fees to the attacker, since the validator client does not check the payload before signing. Cause missed or incorrect attestations. Teku's slashing protection still blocks slashable signatures. | Medium. A validator client using a remote beacon node over the internet is a common setup. |

| Proposer config URL | Replace fee recipients and builder settings for every validator, and change them every epoch if refresh is enabled. | Low. Loading proposer config from a remote URL is uncommon. |

| Builder or relay | Serve a block that pays the attacker, or withhold the payload after the block is signed, causing a missed proposal. Censor transactions. | Low. Only when `--builder-endpoint` points directly at a relay; a local mev-boost checks relay keys itself. |

| Remote execution client | Report invalid payloads as valid, and supply the payloads Teku proposes, including their fee recipient. | Very low. The Engine API almost always runs on the same host or a private network. |

| Checkpoint state, network config, bootnodes | Start the node from an attacker-chosen state or peer set. On its own this mostly leaves the node stuck; with control of its peers too, it can be kept on a fake chain. | Medium, but only at startup. Checkpoint sync over HTTPS is common and is the default on testnets, but it is only used when Teku starts with an empty database. |

| Metrics publishing | Read node metrics and capture any API key in the URL. | Low, and low impact. |

The external signer is the highest-risk endpoint. If your signer requires a client certificate from Teku (mutual TLS), an attacker cannot relay requests to it, which limits the impact to a denial of service.

These settings protect a connection whatever roots Java trusts.

### External signer: pin the signer's certificate

- Set `--validators-external-signer-truststore` and `--validators-external-signer-truststore-password-file` to a trust store holding only your signer's CA. Teku then trusts nothing else for that connection.

- Set `--validators-external-signer-keystore` and its password file so Teku presents a client certificate, and configure your signer to accept only known client certificates.

- Do not put credentials in `--validators-external-signer-url`.

### Startup data and proposer config: use local files

Download the files yourself and pass local paths:

- Checkpoint state: download a finalized state from a source you trust and pass it with `--initial-state` instead of using `--checkpoint-sync-url`.

- Bootnodes and static peers: pass local files to `--p2p-discovery-bootnodes-url` and `--p2p-static-peers-url`.

- Proposer config: pass a local file to `--validators-proposer-config`.

### Remote beacon node, execution client and builder: keep them on a private network

Teku has no option to pin certificates for these connections. Run them on the same host or a private network, and point `--builder-endpoint` at a local mev-boost rather than directly at relays.

We recommend you follow the official guidance for checking whether your Java runtime is vulnerable and fixing it:

- [OpenJDK guidance](

- [OpenJDK vulnerability advisory, 2026-10-21]( CVE-2026-91403

- [OpenJDK vulnerability advisory, 2026-10-24]( CVE-2026-114820

- [Oracle guidance](

- [Oracle Security Alert Advisory: CVE-2026-91403](

- [Oracle Security Alert Advisory: CVE-2026-114820](

If you run a different vendor's Java build, check that vendor's release notes for these fixes.

**Java 26 is end of life.** It was a short-term release and stopped receiving updates when Java 27 shipped in September 2026, so no fixed Java 26 build is coming. If you run Teku on Java 26, move to Java 25, the current long-term support release. This includes the default `consensys/teku` Docker tags such as `latest` and `26.10.2`, which run Java 26: switch to the matching `-jdk25` tag, for example `consensys/teku:26.10.2-jdk25`.

The Teku team is monitoring the situation and can provide further guidance on Discord.

For Docker users, look out for a future release whose images ship without the emSign ECC Root CA - G3. Until then, switch from the default Java 26 tags to the `-jdk25` tags.

If you run Teku from the binary distribution, it uses the Java runtime you installed. Follow the guidance under How to fix your Java runtime to update it.

Thanks to Consensys Diligence, our security partner, for flagging this issue.

- [OpenJDK: Compromised Certificates and OpenJDK Java Runtimes](

- [OpenJDK vulnerability advisory, 2026-10-21]( CVE-2026-91403

- [OpenJDK vulnerability advisory, 2026-10-24]( CVE-2026-114820

- [OpenJDK CheckCerts.java diagnostic](

- [Oracle: guidance on the eMudhra certificates](

- [Oracle Security Alert Advisory: CVE-2026-91403](

- [Oracle Security Alert Advisory: CVE-2026-114820](

- [Mozilla bug 2083449]( the incident record

- [eMudhra statement on the incident](

- [BellSoft: distrusting the eMudhra intermediates in Java]( test results for Temurin and other Java runtimes

- [Azul: workaround for the eMudhra CA certificate compromise](

- [Eclipse Temurin trust store update](

- [Ubuntu: eMudhra CA compromise fixes available]( fixed `ca-certificates` versions for each Ubuntu release, and the Java trust store rebuild

- [Debian bug 1069251]( why ca-certificates-java updates don't refresh Java's trust store

Extracted Entities