All posts

Node.js’s October patches: ten bugs, one that can mean remote code execution.

Node.js 22.23.4, 24.22.1 and 26.12.1 fix ten bugs, including a Brotli flaw that can mean remote code execution. What to check and do now.

Published
Reading time
9 min
By
Girish Babu GV

What happened#

On 21 October 2026 the Node.js project published its October security release, fixing ten vulnerabilities across the three supported lines: 26.x, 24.x and 22.x. Four are rated High, three Medium, and three Low. The fixed versions are 26.12.1, 24.22.1 and 22.23.4. The advisory lists no workaround for any of the ten issues, and it says end-of-life versions are always affected when a security release ships. The release also updates bundled dependencies: c-ares, Undici, the CA certificate bundle (after the compromise of an eMudhra intermediate CA), OpenSSL on 24.x and 22.x, and SQLite on 22.x.

Several of the ten are, by the advisory’s own account, follow-ups to earlier fixes: three are incomplete fixes of CVE-2026-48930, CVE-2026-48931 and CVE-2026-21715, and a fourth, CVE-2026-105483, is a regression introduced after the fix for CVE-2026-48618. That pattern, a security fix that closes one path but leaves a related one open, is common enough in C and C++ runtimes that it is worth remembering next time a bug in this list looks familiar.

The ten issues#

Here is every CVE from the release, in the advisory’s own order, grouped by severity.

All ten CVEs in the October 2026 Node.js security release
CVESeverityAffected linesWhat it is
CVE-2026-111497High26.x, 24.x, 22.xOut-of-bounds write in the Brotli decoder, reachable when an app decompresses Brotli data with node:zlib. Can lead to a crash, memory disclosure or remote code execution.
CVE-2026-105482High26.x, 24.x, 22.xdns.resolve*() and dns.reverse() silently truncate a hostname at a NUL byte, so a name like attacker.example\0.trusted.example resolves as attacker.example.
CVE-2026-105483High26.x, 24.xtls.checkServerIdentity() doesn’t recognise fullwidth-digit IP addresses as IPs, so a certificate with a DNS name that spells out the IP can be accepted for an IP endpoint. A regression after the CVE-2026-48618 fix.
CVE-2026-106069High26.x, 24.x, 22.xLeading-zero IPv4 literals (0127.0.0.1) are read as octal by the WHATWG URL parser but as decimal by some OS resolvers, so a TLS check can validate against the wrong address.
CVE-2026-105481Medium26.x, 24.x, 22.xA reused keep-alive socket can accept a response that arrived before the matching request was sent.
CVE-2026-106072Medium26.x, 24.x, 22.xA DNS-only certificate can authenticate an IP endpoint when the host is a non-canonical IPv4 spelling (0177.0.0.1, 0x7f.0.0.1, 2130706433, 127.1).
CVE-2026-106074Medium26.x, 24.x, 22.xEntropy caches for randomUUID(), randomUUIDv7() and randomInt() are saved into startup snapshots and snapshot-based single executable apps, so every process started from one repeats the same values until the cache refills.
CVE-2026-106070Low26.xzlib.ZipFile can return attacker-controlled bytes after its header validation fails.
CVE-2026-106071Low26.xA wildcard --allow-fs-read or --allow-fs-write rule can over-grant access to a shorter, unintended path.
CVE-2026-106073Low26.x, 24.x, 22.xfs.realpathSync() skips Permission Model read checks, so code running under --permission can still resolve symlinks and enumerate paths it shouldn’t reach.

The one to take seriously: the Brotli decoder (CVE-2026-111497)

The advisory explains that an invalid block switch was reported as “need more input” instead of a format error, so the streaming decoder kept appending bytes to an eight-byte internal buffer. The attack vector it names is Brotli data that an application decompresses with node:zlib. It is the only issue in the release that the advisory says can lead to remote code execution.

A cluster worth knowing about: three ways an IP address can lie to TLS

CVE-2026-105483, CVE-2026-106069 and CVE-2026-106072 are not the same bug, but they’re the same shape of problem: Node.js, the OS resolver and the certificate’s Subject Alternative Name can each read the same string as a different IP address, or fail to recognise it as an IP address at all. A certificate issued for one address can then be accepted for a connection to a different one. Exploiting them has preconditions: for CVE-2026-106072, the advisory says an attacker needs a certificate with a numeric DNS name from a CA the client trusts, plus a platform resolver (such as glibc) that accepts those spellings. CVE-2026-106069 depends on how the platform resolver reads leading zeros (the advisory’s example is macOS). They’re still a reminder that “is this string an IP address” is a harder question than it looks.

Am I affected?#

Unlike a framework bug, this isn’t about which package you installed. It’s about what your code, or a dependency, asks the Node.js runtime to do.

What each class of bug needs to be reachable
If your app or a dependency...Relevant CVEsWorth checking
Decompresses Brotli-encoded request or response bodies with node:zlibCVE-2026-111497Yes, this is the one to prioritise.
Does its own hostname or IP matching for security decisions (allow-lists, SSRF guards, TLS pinning) on top of Node’s dns or TLS modulesCVE-2026-105482, -105483, -106069, -106072Check how you parse and compare addresses, not just whether you use TLS.
Runs long-lived HTTP keep-alive connections to a server you don’t fully trustCVE-2026-105481Lower priority unless that server is adversarial.
Builds a startup snapshot or a single executable app and calls randomUUID() or randomInt() during that build stepCVE-2026-106074Check your build script, not your running app.
Reads untrusted .zip files with zlib.ZipFile, or uses the Permission Model (--permission, --allow-fs-*)CVE-2026-106070, -106071, -106073Niche unless you build on either feature directly.

Most apps that simply run on Node.js, without doing any of the above themselves, are still exposed through dependencies that do. The practical answer for almost everyone is the same regardless: there’s no workaround, so the fix is the same whether or not you can point to the exact line that’s reachable.

What to do now#

  1. Find out what you’re running, everywhere: your own machine, CI, every server, every container image.
  2. Upgrade to 22.23.4, 24.22.1 or 26.12.1, matching your current major line. There is no workaround for any of the ten issues.
  3. Rebuild, don’t just restart, anything that bundles its own Node.js. A container image keeps the runtime it was built with, so it has to be rebuilt from a patched base image. If you ship a snapshot-based single executable app, rebuild that too.
  4. Drop Node.js 20 and older if you haven’t already. They are end-of-life and get no fix for any of these ten issues, or for whatever comes next.
  5. If you do your own IP or hostname matching for anything security-relevant, read the three TLS-adjacent CVEs above and check whether your comparison logic normalises addresses the same way your TLS stack does.
Check, upgrade, verify
# 1. What am I running, right now?
node -v

# 2. Upgrade with nvm (pick your line)
nvm install 26.12.1 && nvm alias default 26.12.1
nvm install 24.22.1
nvm install 22.23.4

# 3. Rebuilding a container? Bump the base image tag, then:
docker build --no-cache -t your-app .
docker run --rm your-app node -v

If you deploy on Vercel#

This site deploys on Vercel, so I checked what that means here specifically, rather than assuming. Vercel’s docs say only major versions (currently 24.x, the default, and 22.x) can be chosen, in Project Settings or with an engines.node range in package.json, and that Vercel rolls out minor and patch updates itself when needed, for example to fix a security issue. This project’s package.json has no engines field, so it takes whatever major version the project is set to, and the minor and patch version within that major is Vercel’s to manage, not mine.

That’s a genuine difference from a self-hosted server: on Vercel, picking up the patch for your major version is Vercel’s job. The docs don’t say when a given patch lands, or whether an existing deployment picks it up without a new build, so check rather than assume: Vercel suggests running node -v in the Build Command or logging process.version.

One real action item, separate from this release: Vercel deprecated Node.js 20 for Builds and Functions on 1 October 2026, which has already passed. If a project is still set to 20 for Functions, Vercel says creating a new deployment shows an error (existing deployments keep working), which is its own reason to move to 22 or 24 if you haven’t.

What I actually checked, for accuracy: locally, where I run the dev server, node -v reports v22.18.0, inside the affected 22.x range and below the fixed 22.23.4. This repository has no engines field, vercel.json or vercel.ts, so it takes Vercel’s project-level default. I also searched this codebase for direct use of the lower-level APIs these CVEs touch, Brotli or zlib decompression, dns.resolve, fs.realpathSync, randomUUID, randomInt, or the Permission Model flags, and found none: this site doesn’t call any of them itself. I didn’t redeploy to confirm the production runtime’s exact patch version for this post, since that’s a one-line check I can do at the next real deploy, not something worth a special release for a site with no code path into any of these ten bugs.

Lessons for developers#

My own take, not from the advisory.

  • A platform can absorb a patch for you, but only for what it manages. Vercel patches the Node.js minor and patch version. It can’t patch a Docker image you build yourself, or a dependency with its own vendored copy of a vulnerable library. Know which category each of your deployments falls into.
  • “Is this string an IP address” is a security question, not a parsing detail. Three of the ten bugs exist because different layers of the same stack, the resolver, the URL parser and the certificate check, can disagree about what a string represents. If you compare addresses yourself anywhere, use the same normalisation your TLS library does, or don’t compare strings at all.
  • An “incomplete fix” is still a fix you need. Three of this release’s ten bugs are gaps left by earlier patches. That’s not a reason to distrust security releases, it’s a reason to keep applying them.
  • End-of-life means every future bug too, not just today’s. Node.js 20 and older don’t get a fix for any of these ten, and won’t get a fix for whatever the November release covers either.

Sources#

Keep reading

Esc