Node.js fixes 12 HTTP/2 and TLS flaws in November release
Node.js fixes 12 vulnerabilities across HTTP/2, TLS and stream handling. Learn which 22, 24 and 26 releases to deploy and how to regression-test them.
- AUTHOR
- Karol Rapacz / CEO of Breachroad · OSCP · PNPT
- PUBLISHED
- 11 November 2026
- READING TIME
- 20 min read
- TOPIC
- Vulnerabilities and CVEs
The Node.js project released security updates for supported lines on 11 November 2026. Versions 22.23.5, 24.22.3 and 26.13.1 fix a group of vulnerabilities across HTTP/2, TLS, stream handling and dependencies. Several High-severity flaws can cause remote process termination, memory disclosure or bypass of client-authentication assumptions.
The official Node.js November 2026 Security Releases advisory is the primary source for versions, CVEs, affected branches and technical descriptions. We do not include exploit code; the focus is how application owners should assess reachability and deliver the change safely.
The most important vulnerabilities
CVE-2026-123863 (High) is a remotely reachable HTTP/2 assertion failure during abort handling, affecting Node.js 24 and 26. Where a process receives untrusted HTTP/2 traffic without an intermediary that actually terminates and normalises it, the issue can affect availability.
CVE-2026-123864 (High) concerns HTTPS reentrancy during a zero-length TLS write and can disclose heap memory. That memory may include request fragments or TLS keys. Lines 22, 24 and 26 are affected.
CVE-2026-123865 (High) is an HTTP/2 protocol error during shutdown that can leak sessions, memory and file descriptors. A long-running process may lose resources under remotely supplied traffic. It affects 22, 24 and 26.
CVE-2026-123866 (High) describes a use-after-free during stream teardown and possible response-memory disclosure. Supported lines 22, 24 and 26 are affected.
CVE-2026-126051 (High) involves deferred TLS callbacks and a remotely triggerable uncaught exception in lines 24 and 26. A process may terminate unless the application isolates worker failure.
CVE-2026-123872 (High) concerns TLS 1.2 renegotiation and a potential bypass of per-host client-certificate authentication across lines 22, 24 and 26. It is particularly relevant where mTLS is the authorisation boundary between tenants or services.
CVE-2026-123874 (High) is a use-after-free in http2.respondWithFile, affecting lines 24 and 26. Exposure depends on whether an application or framework reaches that code path.
The advisory also contains Medium and Low issues and updates dependencies including llhttp, undici and sqlite. Do not cherry-pick the CVE with the highest score: the new runtime provides a coherent fix set.
Does a reverse proxy remove exposure?
It depends on the architecture. If a CDN or proxy terminates HTTP/2 and sends only HTTP/1.1 to Node, some runtime HTTP/2 paths may not be Internet-reachable. That does not protect internal services speaking HTTP/2 directly or libraries that create servers on separate ports. A proxy also does not remove TLS flaws when Node terminates TLS itself.
Confirm reachability through configuration and telemetry. Check protocol on every hop, health-check ports, administrative listeners, sidecars and direct partner connections. Documentation diagrams can lag reality; logs and connection tests show the effective state.
What to update first
Prioritise public Node servers handling HTTP/2 or TLS, API gateways, mTLS applications, multi-tenant services and long-lived processes. Then update internal services reachable from user and CI segments.
Functions such as respondWithFile may require source and dependency searches. The absence of a direct call in the repository does not prove safety because a framework may use it internally. SBOM ownership should determine the runtime in the deployed image, not only the engines declaration in package.json.
A safe update plan
Build images using a fixed runtime and reinstall dependencies from the lockfile. Do not copy node_modules between different ABIs or base images. Verify the image signature or digest and create a fresh SBOM.
In test, exercise:
- HTTP/1.1 and HTTP/2, including cancellation and connection shutdown;
- large and zero-length TLS responses;
- upload, streaming, Server-Sent Events and WebSocket where used;
- mTLS with valid, invalid and missing certificates;
- reconnection to databases, queues and external APIs;
- static files and range requests;
- memory and descriptor limits plus worker failure behaviour.
Deploy a canary and observe errors, p95/p99 latency, RSS memory, HTTP/2 session count, descriptors, restarts and TLS status. Increase traffic gradually. Retain rollback capability, but avoid blindly returning to an affected build; prefer fixing the regression or constraining traffic.
Containers and serverless
The Node version in a developer shell does not reveal production state. Runtime may come from a base image, managed platform, buildpack or serverless layer. Check the digest of the deployed artefact and effective version in every region. A moving node:24 tag is not reproducible evidence.
In serverless, a provider may patch a managed runtime, but the organisation still needs the schedule and may need to redeploy. Zero-traffic functions can be invoked later. Remove old versions and aliases that still route a share of requests to an affected environment.
mTLS needs specific verification
CVE-2026-123872 affects the assumption that a client certificate was validated for the correct host and context. In multi-host architecture, check SNI, virtual hosts, session caches, TLS 1.2 renegotiation and certificate policy. An authorized: true result without binding to the expected tenant can be insufficient.
In addition to upgrading, enforce TLS 1.3 where compatibility permits, narrow trusted CAs and bind certificate identity to a service. A negative test should present a valid certificate for the wrong context and confirm rejection. API security still requires application authorisation when transport uses mTLS.
Detection and hunting
Historical review should look for unexplained process restarts, assertion failures, HTTP/2 errors, growing memory or descriptor counts and unusual renegotiation. One crash does not prove an attack. Correlation with source, frame sequence, repetition and traffic to multiple instances raises confidence.
Do not upload full memory dumps to a shared system without control. A core file may contain keys and user data. Treat it as sensitive evidence with restricted access and retention. If TLS-key disclosure is plausible, rotate the key after closing the affected path.
Supply-chain ownership
Runtime updates are part of software supply-chain security. The platform team provides a fixed base, application owners run regression tests, and security sets priority and monitors exposure. A central image helps but does not prove every repository uses it.
Our DevSecOps pipeline guide covers artefact gates and SBOMs. A security audit can establish whether production matches declared versions and whether runtime, images and managed services are included in the update process.
Node.js facts and Breachroad conclusions
Source facts: on 11 November the project released 22.23.5, 24.22.3 and 26.13.1; its advisory lists HTTP/2, TLS, use-after-free and resource-leak vulnerabilities plus dependency updates. Breachroad conclusions: service priority, the test set, canary, hunting and mTLS verification are deployment recommendations. Not every application exposes every described path.
What application owners should do
Today, identify every effective Node version, map direct HTTP/2 and TLS exposure and schedule updates. This week, build a new artefact, run protocol tests, canary it and remove stale images. This month, add automated runtime evidence, an SBOM and alerts for unsupported lines.
If you need to test malformed protocol handling, stream cancellation and mTLS safely, a web application and API penetration test can cover it within an authorised scope. The main lesson is simple: the runtime is part of the application and needs ownership just like business code.
Building an exposure matrix
For every service, record Node version and LTS line, client-to-proxy and proxy-to-application protocols, TLS termination, mTLS use, HTTP/2 APIs and owner. Attach evidence: image digest, runtime output, ingress configuration or a connection trace. “There is a proxy” then stops being an unverified assumption.
Include developer tools, webhooks, panels and jobs, not only the main service. A helper process in the same image may expose a port outside standard ingress. Prioritise by reachability and impact, but every supported workload should receive the fixed runtime.
Resilience tests beyond functionality
Regression should detect memory and descriptor growth across repeated session open and close. Establish a baseline, run valid and invalid sequences, then verify resources return to the expected level. A limit without observation may only postpone failure.
In multi-process deployments, test load-balancer behaviour after worker crash, request loss, retry and idempotence. Automatic restart improves availability but can hide exploitation. Alerts should join restart, protocol error and traffic source.
TLS keys and possible disclosure
If an affected build and CVE-2026-123864 path were reachable and telemetry is anomalous, assess key and certificate rotation. Reissuing a certificate without changing its private key does not close exposure. Review session tickets and every copy of the key.
Rotate after deploying the fix so new material does not enter the same affected process. For a key shared across services, map every consumer and avoid uncontrolled outage. Long term, use separate keys and a managed store.
Maintaining LTS lines
Platform owners should publish end-of-support dates and automatically block new deployment on an obsolete line. Exceptions need a dependency, migration plan and expiry. A patch update within a supported line should be routine rather than a weeks-long project.
Rebuild images from scratch monthly and clear caches to reveal hidden dependencies. A pinned digest is reproducible but needs regular refresh; a moving tag is current but weakens deployment evidence. A mature pipeline combines controlled updating with immutable artefacts.
Post-deployment metrics
Measure advisory-to-image time, services with version evidence, protocol-test coverage, exceptions and closure time. For at least a week, compare restarts, memory, descriptors and TLS errors with baseline. No regression and removal of old registry images are both completion criteria.
Share findings through concise DevSecOps security training so the next runtime update starts from a tested matrix rather than rebuilding the process.


