Node.js October 2026: A Brotli RCE and Three Ways to Accept the Wrong Certificate
Node.js patched ten vulnerabilities on 21 October 2026, including an out-of-bounds write in its bundled Brotli decoder and three separate ways for TLS hostname verification to accept a certificate issued for a different address.

What shipped
On 21 October 2026 the Node.js project released security updates for its three supported lines, fixing ten vulnerabilities. The patched versions are 26.12.1, 24.22.1 and 22.23.4. Four of the ten are rated High. The headline issue is CVE-2026-111497, an out-of-bounds write in the Brotli decoder bundled with Node.js. The advisory is explicit that impact ranges from a crash through process memory disclosure to remote code execution. It was reported by OpenAI Codex Security.
The Brotli flaw, and why it reaches further than it looks
The decoder mishandles an invalid block switch: it treats the malformed input as a signal that more data is needed, and continues writing past the end of an internal eight-byte buffer. Any application that decompresses untrusted data through `node:zlib` is in scope. That is a wider surface than most teams assume. Brotli decompression is not an exotic code path — it is what happens when a service accepts a `Content-Encoding: br` request body, when a server-side HTTP client fetches a compressed response from a third-party API, or when a background job unpacks an uploaded artefact. The attacker does not need an account, only a route that accepts compressed input. Two details are worth holding onto. First, researchers demonstrated code execution against an unmodified Node.js build under laboratory conditions, and the exploit did not succeed on every attempt — unreliable exploitation is still exploitation, and it is not a reason to deprioritise. Second, and more importantly for mixed estates: this is a Brotli vulnerability, not a Node vulnerability. PHP patched the same CVE in updated Windows builds on 26 October, because both runtimes bundle the same library. If you run PHP and Node side by side, patching one does nothing for the other, and a version check against the wrong runtime will tell you everything is fine when it is not.
Three ways to validate a certificate against the wrong address
The more subtle group in this release concerns `tls.checkServerIdentity()`, where three separate flaws all end in the same place: a certificate accepted for an IP it was never issued for. CVE-2026-105483 (High) is a regression. Hostname verification does not recognise IP addresses written with fullwidth digits — `127.0.0.1` rather than `127.0.0.1` — so the value is matched as a DNS name instead of an IP, and a certificate carrying `DNS:127.0.0.1` is accepted. CVE-2026-106069 (High) exploits a disagreement about number bases. WHATWG URL normalisation reads a leading-zero IPv4 literal such as `0127.0.0.1` as octal, while some operating system resolvers — macOS among them — read it as decimal. The certificate is therefore validated against a different address than the one the socket actually connects to. CVE-2026-106072 (Medium) is the same theme in other spellings: `0177.0.0.1`, `0x7f.0.0.1`, `2130706433` and `127.1` can all cause a DNS-only subject alternative name to authenticate an IP endpoint. Individually these look like edge cases. Together they describe a pattern worth internalising: whenever two components parse the same string with different rules, the gap between them is a vulnerability. Two of the three are explicitly regressions from earlier fixes, which is a reminder that a CVE being "already patched once" is not assurance.
The quietly dangerous one
CVE-2026-106074 is rated Medium and deserves more attention than its rating suggests. The entropy caches behind `crypto.randomUUID()`, `randomUUIDv7()` and `randomInt()` are captured into startup snapshots and single executable applications. Every process started from that snapshot then produces the same sequence of values. If you use startup snapshots or SEAs — increasingly common for fast cold starts in serverless and containerised deployments — then identifiers you believed were unique and unguessable may be neither. Anything built on those values deserves review: session tokens, password reset links, idempotency keys, object identifiers. This is the kind of flaw that does not announce itself in logs and surfaces much later as a collision nobody can explain.
End-of-life lines
Node.js 20.x, 23.x and 25.x are affected and will not be patched. As with PHP's unsupported branches, the only remedy is moving to a supported line. Teams running containers built on an older base image should check what is actually inside the image rather than what the Dockerfile says, since a pinned base can lag well behind.
What to do
Upgrade to 26.12.1, 24.22.1 or 22.23.4. Then check your container images, CI runners and serverless runtimes separately — these are routinely missed because nobody thinks of them as servers, and a build pipeline that decompresses untrusted artefacts is squarely in scope for the Brotli issue. If you use startup snapshots or single executable applications, treat CVE-2026-106074 as a data problem as well as a patching one, and review anything generated from the affected random functions. Finally, if your estate includes PHP, close the Brotli issue on both runtimes. One library, two patch processes.
How IIT Oman can help
Our DevOps and cloud engineering practice builds the pipelines that make this routine rather than disruptive: dependency and base-image scanning, automated patch workflows, and staged rollouts with a rollback path, so a critical advisory becomes a planned deployment instead of an emergency. Where applications are stranded on an end-of-life Node line, we carry out upgrade assessments and incremental migrations that keep the service running while the runtime moves underneath it. Our cybersecurity practice runs VAPT and compromise assessment where an exposure window existed, and we can review TLS and certificate handling across service-to-service calls, which is exactly where the hostname verification flaws in this release would bite. If you are not certain which Node versions are running across your production, CI and container estate, we can establish that for you. Talk to our engineering team in Muscat.

