On 11 November 2026 the Node.js project published security releases 26.13.1 'Lithium' (LTS), 24.22.3 'Krypton' (LTS) and 22.23.5 'Jod' (LTS). If you run Node.js on the 22.x, 24.x or 26.x lines and have not upgraded since then, treat yourself as exposed: the fixes cover a remotely triggerable TLS server crash, a client-certificate authentication bypass, and HTTP/2 memory-safety bugs that can leak heap memory belonging to other connections. Upgrading to the build for your release line and restarting is the entire fix — the advisory offers no configuration workaround.
What the November 2026 Node.js security release fixes
This is a core security release, meaning the defects live in Node.js itself, not in a package you chose to install. The release notes tag each entry with a CVE identifier and a severity, and nearly all of the high-severity items sit in two modules: http2 (stream lifecycle, buffer retention, session memory accounting) and tls (resumption, renegotiation, SNI handling, deferred handshake callbacks). The remaining items are a handful of medium and low fixes across the permission model, zlib, and credential redaction in HTTP proxy error messages, plus a dependency bump of llhttp to 9.4.4. All three LTS lines were published within seconds of each other on 11 November, so there is no line where you are waiting on a backport.
Do these flaws affect me?
If your Node.js process terminates TLS or serves HTTP/2 to anything beyond a trusted internal caller, assume yes. The vendor write-up lists the HTTPS re-entrancy flaw as affecting "all users in active release lines: 22.x, 24.x, 26.x", and the HTTP/2 stream-cleanup flaw as affecting 24.x and 26.x. Words like "unauthenticated" and "remotely triggerable" appear repeatedly, which is the tell that a public-facing endpoint is the exposure to plan for.
One assumption worth testing: teams that terminate TLS at a load balancer or reverse proxy often conclude the TLS flaws cannot reach them. That may hold for handshake-level issues, but CVE-2026-123874 is described as reachable "over both TLS and cleartext", so an internal-only HTTP/2 hop is not automatically safe. If you run Node behind nginx and only speak HTTP/1.1 to the app, your surface is smaller — not zero.
The flaws worth prioritising first
Start with the two that break confidentiality and authentication, not the ones that break availability.
- CVE-2026-123864 (tls, http) — a zero-length TLS write re-enters llhttpexecute(), and the server copies its own heap into the request body. Per the advisory, an unauthenticated client can read memory belonging to other connections, including TLS private keys and other users' requests.
- CVE-2026-123872 (tls) — a client certificate authentication bypass: a client can change the servername during TLS 1.2 renegotiation to reach endpoints it was not authorised for. If you use client certificates to separate tenants or services, read this one twice.
- CVE-2026-126051 (tls) — a deferred-callback use-after-destroy in TLS server handshake handling that lets a remote, unauthenticated attacker terminate the server process.
- CVE-2026-123865, CVE-2026-123866 and CVE-2026-123874 (http2) — a permanent memory and file-descriptor leak when a protocol error hits during connection shutdown, a use-after-free that can send freed response memory to the peer, and a use-after-free when a respondWithFile() stream is reset during a pending write.
What to do in the next 24 hours
The action is a patch upgrade and a restart: 26.13.1, 24.22.3 or 22.23.5, whichever matches your release line. Find every place a Node version is pinned — the Dockerfile base image, a digest in a Helm chart, .nvmrc, package.json engines, CI build images, serverless runtime settings — and change it there. Long-lived processes started before 11 November are still unpatched even if the image tag looks current.
Then verify what is actually serving traffic, not what is in the repository. Expose process.version on a health endpoint and curl it through the load balancer, or run node -v inside the running container on each host. Confirm the running container's image digest, not just its tag, and drain old pods rather than trusting the rollout status. This is the same runtime-hygiene work we do under server and runtime work; the pattern repeats in almost every dependency incident, because the gap is rarely the upgrade and almost always the verification.
What to watch in the logs afterwards
Expect nothing dramatic. The value is in anomalies you would otherwise dismiss as noise.
- TLS handshake failures and unexpected SNI mismatches, especially from clients presenting certificates.
- HTTP/2 GOAWAY frames and stream resets that do not track request volume.
- Resident memory that climbs and does not come back down, and file descriptor counts creeping toward the limit (EMFILE) — the signature the advisory describes for CVE-2026-123865.
- TLS server processes dying with no application stack trace, which is how CVE-2026-126051 presents.
Give it a week of watching. Two of these bugs are leaks that accumulate rather than fail loudly, so a quiet first hour proves very little.
What AI-assisted review changes about this class of bug
The advisory credit is the interesting part: it names Chris Rohlf, in collaboration with Claude and Anthropic Research, on the HTTPS re-entrancy flaw (CVE-2026-123864) and the HTTP/2 shutdown leak (CVE-2026-123865). That matters less as a headline than as a description of how memory-safety bugs get found now.
Notice the shape of two entries. CVE-2026-123865 is an incomplete fix of CVE-2026-48937, and CVE-2026-123863 is a variant of CVE-2024-27983 — the earlier patch "only detached streams still tracked in the session". Most core bugs of this class are recurrences: one ownership assumption, missed at one call site. That is exactly the sweep a model-assisted review is good at once you hand it a patch diff and ask where the same assumption appears again. The practical workflow for your own code: read the patch, ask for other places where a callback can fire after destroy or where a buffer outlives its stream, then verify each candidate in a debugger or an instrumented build. Treat the output as candidate test cases, not findings. The false positives are confident and plausible, and someone still has to understand the event loop.
Where to start
Patch tonight, verify the version answering requests on every host, then watch memory, descriptors and handshake errors for a week. If your Node estate is spread across containers, a couple of long-lived VMs and a serverless runtime or two, the sequencing is usually the hard part rather than the upgrade itself — Maysan Labs does this kind of runtime work regularly, and you can start here if you want a second pair of eyes on the rollout.
Buying software or a website?
Twelve questions to ask any vendor before you sign: price, your data, fit for an Indian business and support.
Get the free checklistWritten by Maysan Engineering Team at Maysan Labs, a software studio in Gurgaon, India. Follow us on LinkedIn