The patch wave: .NET, Node.js, Brotli and Ubuntu in one week
Microsoft shipped out-of-band .NET updates on 25 October. Node.js released 22.23.4, 24.22.1 and 26.12.1 on 21 October, Google’s Brotli project tagged v1.2.1 the same day, and Ubuntu issued USN-8956-1 on 26 October. All of them address CVE-2026-111497, an out-of-bounds write in the Brotli decompressor that an attacker can reach through data an application decompresses.
The code change that prevents the flaw has existed since 27 February 2026. It reached no upstream release until 21 October. That gap is why the patch wave landed in the last week of October, and why the work is spread across language runtimes, API gateways, proxies and vendored copies of a library that most of these projects bundle rather than install by hand.
A fix committed on 27 February reached no release until October
The upstream change is commit 0ed4268, “Detect and report invalid stream earlier.”, committed at 2026-02-27T13:10:32Z by Evgenii Kliuchnikov. It changes DecodeBlockTypeAndLength to return a BrotliDecoderErrorCode, replaces a reserved -17 value with BROTLI_DECODER_ERROR_FORMAT_BLOCK_SWITCH, and adds a BROTLI_SAFE_WITH_STATUS macro at the three block-switch call sites. The February date is taken from the commit page itself.
The Register quotes OpenAI’s account of the gap: the root cause “was actually fixed in the Brotli source code in February, but it wasn’t known at the time that it gave rise to a security vulnerability, and there was no more recent release of Brotli.” OpenAI’s own page returned HTTP 403 when checked for this article, so the quotation is carried as reported by The Register.
Disclosure ran through the October releases. Node.js listed the Brotli item as HIGH in its October security release and recorded a cherry-pick of 0ed4268. google/brotli v1.2.1 was tagged on 21 October with a SECURITY note naming CVE-2026-111497. NVD and the GitHub advisory published on 22 October, the same day Imperva published its Node2Shell demonstration, and PYSEC-2026-4322 followed on 23 October. Microsoft’s batch arrived on 25 October, then Ubuntu’s USN-8956-1 and The Register’s report on 26 October.
OpenSourceMalware records the flaw as disclosed on 21 October by Imperva, while Imperva’s own post is dated 22 October and dates the disclosure to the Node.js release the day before.
Why a merged fix is not a shipped fix
Brotli sits inside other software in several ways. It can be statically linked, vendored into a source tree, bundled into a runtime pack or container image, or pinned in a lockfile. Each of those copies has to be updated by its own project, and three examples from the past week show how slowly that happens.
dotnet/runtime issue #136422 was opened on 24 October by a community reporter reading the vendored decoders in 9.0.21 and 10.0.13; the reporter did not claim to have confirmed exploitability. A Microsoft member replied the same day and posted the list of fixed builds (8.0.33, 9.0.22, 10.0.14, 11.0.0-rc.3) on 25 October, closed the issue, and reopened it the same day “to track porting the fixes from our internal release branches into the public GitHub branches”. It was still open at the time of writing, so builds taken from the public branches may remain affected.
ClickHouse PR #127419 bumped Brotli from 1.2.0 to 1.2.1, merged into master on 25 October and backported to several maintenance branches, with a regression test asserting that the decoder returns -9 on 1.2.0 and -17 on 1.2.1. JFrog reported that no patched ClickHouse release had shipped when it published.
apache/arrow issue #55166, opened on 23 October, records that the project bundles Brotli 1.0.9. Builds that link a patched system Brotli are unaffected, and a pull request moving the bundled version to 1.0.10 was approved but unmerged, according to the tracker.
JFrog’s summary of the pattern is the clearest one available: “fixing the upstream decoder does not automatically make every affected application safe: downstream projects must incorporate the fix, release an updated version, and ensure that their distributions or deployments use it.”
The affected upstream range is Brotli 1.0.3 through 1.2.0, fixed in 1.0.10, 1.1.1 and 1.2.1, according to the GitHub advisory, Ubuntu and PYSEC records. A patched decoder defines BROTLI_DECODER_ERROR_FORMAT_BLOCK_SWITCH, the -17 code that the February commit introduced. Wiz’s affected-and-patched table lists patched releases for Envoy (1.40.0) and Istio (1.29.9, 1.30.6 and 1.31.2), which is why the story reaches API gateways and service meshes as well as runtimes.
The wider affected-project list comes from the broccoli2shell.com tracker, which is independent and says so: it is not affiliated with Google, Node.js or the Python Software Foundation, and it warns that its list is incomplete because “GitHub’s code search skips submodules and some very large repositories, and private or vendored-at-build-time copies don’t show up at all”.
Microsoft’s advisory: 8.1, CWE-787, and no in-the-wild claim
Microsoft’s MSRC update-guide page for the CVE rendered as an empty shell, so Microsoft’s claims here come from the .NET blog post and the text of the dotnet/announcements issue #494.
The advisory text: “A buffer overflow vulnerability exists in .NET’s use of the Brotli decompression algorithm. This vulnerability may allow an unauthenticated attacker to read or corrupt process memory. This could lead to information disclosure, denial of service, or elevation of privilege.” Microsoft rates it CVSS v3.1 High, 8.1, with CWE-787 (out-of-bounds write) and all platforms and architectures affected.
Microsoft does not claim exploitation in the wild, and its vector carries E:U, the unexploited sub-score. The “Exploitation More Likely” assessment is attributed by The Register to Microsoft’s advisory; it could not be verified in the readable Microsoft material, so it belongs to The Register’s reporting.
Severity is not a single number here, and the sources should not be blended. Microsoft says 8.1 under CVSS v3.1. GitHub’s advisory GHSA-9v8h-qqgm-4m35 rates it Critical at 9.2 under CVSS v4.0. Node.js calls it HIGH, and PYSEC-2026-4322 carries the v4 vector from GitHub. The main independent check is Ubuntu’s Medium priority, recorded in Ubuntu’s CVE entry on 23 October and fixed in USN-8956-1 on 26 October; the oss-security thread singled that out as a sign that OS providers may be under-rating the issue. Ubuntu’s rating is a distribution-level assessment and does not revise Microsoft’s number.
Imperva demonstrated code execution against an unpatched Node.js build, and a public proof of concept is reported to require a deliberately weakened build; Microsoft’s exploitability assessment reaches readers through The Register’s reporting. The tracker states that “as of 2026-10-22, no attacks in the wild had been reported”.
Which code paths actually reach the decoder
.NET applications are exposed when they decompress attacker-controlled Brotli data through ASP.NET Core request decompression, HttpClient automatic response decompression, or the System.IO.Compression Brotli APIs (BrotliStream and BrotliDecoder). The first two are not enabled by default, and Microsoft is explicit that “an application is not exposed merely because an affected .NET package is installed”. NativeAOT and Mono runtime packs and Microsoft.DotNet.ILCompiler are also affected. .NET Framework is not. Microsoft also warns that mitigation snippets should only modify decompression configuration that already exists: adding AddRequestDecompression or an AutomaticDecompression assignment where none was configured can introduce denial-of-service risk.
Node.js’s own advisory names node:zlib, and Imperva’s Node2Shell research demonstrated code execution through Brotli responses consumed by the built-in fetch() and through request bodies inflated by Express 5.0.0 through 5.3.0 and body-parser 2.0.0 through 2.3.0. Imperva is careful about the middleware: a bare node:http server does not decode request bodies by itself, so the parsers mark a reachable path and the flaw itself sits in the decoder.
Python’s exposure runs through the PyPI packages Brotli and brotlicffi, which compile the same c/dec/decode.c. Where either is installed, urllib3 (and therefore requests), httpx with the brotli extra, and aiohttp decode Brotli responses automatically, and inbound Content-Encoding: br bodies are the request-side equivalent. OpenSourceMalware notes that both PyPI releases were patched silently, with no mention of the CVE in their release notes.
Two more implementations, from the tracker’s analysis rather than independent verification: andybalholm/brotli for Go had the same logic error and was fixed on 23 October in v1.2.7, while Rust’s brotli-decompressor detects the condition and returns an error, and the Java, JavaScript and C# ports cannot reach the bug.
JFrog’s own analysis extends the reach further, and it should be read as vendor-extended analysis rather than confirmed scope. Crafted JPEG XL input processed through libjxl can reach Brotli data in FFmpeg builds, and Apache Arrow can be driven into an indefinite CPU loop, a denial of service, from a small input. JFrog itself notes that remote code execution would require further exploitation and depends on the build and the mitigations in place.
How to check your stack and why rebuilds are required
A running process keeps the decoder it loaded until it restarts. A container keeps the decoder baked into its image until the image is rebuilt on a patched base and redeployed. A statically linked or vendored copy ignores a distribution update entirely, and lockfiles and version pins have to be raised or the next build reinstalls the vulnerable version.
The tracker publishes short checks, and one block covers the common stacks:
node --version # patched: 22.23.4, 24.22.1, 26.12.1 or newer
node -p "process.versions.brotli" # the bundled decoder, not just the runtime version
docker run --rm --entrypoint node your-image --version # container images carry their own runtime
pip show Brotli brotlicffi # fixed: 1.2.1 and 1.2.1.0
dotnet --list-runtimes # framework-dependent installs
There is no fix for Node.js’s end-of-life lines, which include 20, 23 and 25: the tracker’s advice is to move onto 22, 24 or 26.
Microsoft’s own instructions separate the cases. Framework-dependent apps install the updated runtime and restart. Self-contained apps must “rebuild and republish with the updated runtime”, because “installing a newer runtime on the host does not update the runtime bundled with your application”. Containers must be rebuilt and redeployed on updated .NET images. NativeAOT is the sharpest case: “Updating the machine’s shared runtime alone does not replace a decoder already linked into a NativeAOT executable.” Self-contained and container deployments need checking separately from the host.
The temporary measures the sources describe are defensive: reject inbound Content-Encoding: br at the proxy or WAF, or remove the Brotli binding where that is practical. Two limits apply. Stripping br from Accept-Encoding does not stop a malicious server from sending Brotli anyway, and a WAF cannot see responses your application fetches. Microsoft’s warning about not adding decompression configuration that was not already there applies to these measures as well.
Scanners missed the Python packages
The clearest evidence of the lag came from the Python side. OpenSourceMalware wrote on 23 October that the CVE record named github.com/google/brotli as the affected product and “does not mention PyPI”, that “as of today no advisory database lists either PyPI package, so your scanners will not tell you”, and that “tools like Dependabot and Safety will not flag this yet”. Its stated basis was searches of OSV, PYSEC and GHSA that returned no record for the Python packages.
The gap closed within days. The PyPI security team designated PYSEC-2026-4322 for brotli and PYSEC-2026-4323 for brotlicffi, and the GitHub advisory was updated to add both packages. The PYSEC record for brotli is dated 23 October, aliases CVE-2026-111497 and lists 1.2.1 as the fixed version. GHSA-9v8h-qqgm-4m35 now lists brotli from 1.0.3 up to 1.2.1 and brotlicffi from 1.0.3 up to 1.2.1.0 as affected, with google/brotli as its source code field, and the advisory page showed “Dependabot alerts 0”. That detail is consistent with no alert having been raised for the Python packages at all.
The sources read for this article name Dependabot and Safety; pip-audit appears only in the tracker’s wording that “pip-audit, Dependabot and Safety didn’t reliably flag” the packages, so that last claim rests on the tracker alone.
Four names, and the team that found it
Searches for this flaw may arrive under any of four names: Node2Shell (Imperva, 22 October, Node-centred), HTTP2Shell (23 October, a broader HTTP framing per JFrog), Broccoli2Shell (Wiz’s headline and the tracker’s domain) and Brotli2Shell (JFrog’s framing). The tracker is blunt about its own contribution: “we didn’t find this bug. We just registered a domain named after a vegetable.” None of the names describes the flaw, which sits in the C decoder.
The Node.js advisory credits OpenAI Codex Security, Khai Tran and Sylvie Mayer for reporting the vulnerability, and mcollina and aduh95 for the fix. Wiz adds Ron Masas of the Imperva Red Team for demonstrating exploitability and OpenSourceMalware and Aikido for the Python impact analysis. Imperva’s own post attributes the report to Matteo Collina and the fix to Antoine du Hamel; the reporting credit conflicts with the Node.js advisory.
The Register notes that Node.js did not explain why the flaw was recognised only in October, and that OpenAI filled the gap in a post on X. Codex Security is OpenAI’s security tooling and the finder of record here. The wider argument about agentic AI tools in open source is a separate discussion; it does not establish anything about how this bug behaved.
The same week’s other .NET problems
Wiz published PHP2Shell on 21 October, describing core PHP engine flaws that allow remote code execution and Heartbleed-style memory disclosure. The Register reports that the Brotli flaw “went slightly under the radar due to the disclosure timing closely aligning with that of PHP2Shell”. Different CVE, different component, and a timing collision rather than a shared cause.
Two .NET advisories were re-released in the same batch: CVE-2026-70197, a SocketAsyncEventArgs memory corruption issue, and CVE-2026-102870, a Blazor WebAssembly debugger issue. Microsoft re-released both because “the October 13, 2026 release did not fully address these issues”. Both are unrelated .NET defects that travelled in the same batch as the Brotli fix.
What is still unclear
The exposure estimates that circulated with the story are Wiz’s own telemetry: 91% of cloud environments with vulnerable resources and 82% with publicly exposed Node.js instances, repeated by The Register with attribution to Wiz’s Merav Bar. Those figures come from Wiz’s own scans of its customers, so they are not an independent census of the internet.
The tracker’s list is incomplete, so a project that does not appear on it has not been cleared.
