• Terms of Service

Learn Internet Grow

Tech Skills

  • Learn
  • Books
  • Search
  • Create
  • Measure
  • Secure
  • Share
  • Terms of Service

Microsoft Patched .NET for a Bug That Isn’t .NET’s. Most Apps Were Never Exposed.

Filed Under: Secure

Microsoft Patched .NET for a Bug That Isn't .NET's. Most Apps Were Never Exposed — a painterly illustration conveying this idea from "Microsoft Patched .NET for a Bug That Isn't .NET's. Most Apps Were Never Exposed."

On 25 October 2026, Microsoft did something it does not do often: it shipped a .NET security update out of band, outside the monthly patch cycle, twelve days after the regular October release. The .NET security update fixes a flaw in the Brotli decompression library, and the coverage that followed used words like critical, emergency and remote code execution.

Here is the part almost nobody led with, in Microsoft’s own words:

“An application is not exposed merely because an affected .NET package is installed.”

Most .NET applications were never in danger, because the feature that carries the risk is switched off by default and you have to go and turn it on. That does not make the update optional. It does mean the honest first step is not panic — it is a two-minute check to find out whether this is your problem at all.

And there is a second thing worth knowing before any of the version numbers: this is not a new vulnerability, and it is not really a .NET one. It is the same flaw we covered in Node.js four days earlier, arriving in its second major runtime.

Everything below was checked against Microsoft’s security advisory, the .NET engineering blog post, the Node.js project’s advisory and Microsoft’s .NET support policy, all read on 26 October 2026.

The thirty-second answer: are you actually exposed?

The thirty-second answer: are you actually exposed? — a painterly illustration conveying this idea from "Microsoft Patched .NET for a Bug That Isn't .NET's. Most Apps Were Never Exposed."

Brotli is a compression format — the same family of idea as ZIP — used constantly and invisibly to make web traffic smaller. The bug is in the code that uncompresses it. So the question is simply: does your application decompress Brotli data that came from somewhere you do not control?

In .NET, there are exactly three ways that happens, and Microsoft lists all three. Every one of them requires somebody to have opted in:

  1. ASP.NET Core request decompression, with the Brotli (br) provider enabled. Off by default.
  2. HttpClient automatic response decompression, where the handler’s AutomaticDecompression setting has been set to include Brotli — or to All, which quietly includes it. Off by default.
  3. Direct use of the Brotli APIs in System.IO.Compression, such as BrotliStream or BrotliDecoder, on data you did not produce.

If none of those describes your code, you are not exposed through .NET, and you can install the update on a normal schedule rather than a frantic one. The .NET blog says it plainly: “Request decompression and automatic response decompression are not enabled by default.”

Two more exclusions worth having:

  • .NET Framework is not affected. The old Windows-only .NET — the 4.x line still running an enormous amount of business software — is explicitly out of scope.
  • Installing a package is not exposure. The advisory says so directly, and it is the sentence that separates a real inventory problem from a scanner result that frightens people for no reason.

⚠️ The number-three case is the one people miss. The first two are framework features with obvious switches. The third is a line of code somebody wrote years ago to unpack an upload or a feed, and nothing in a configuration file will reveal it. If you are doing this properly, search your code for BrotliStream and BrotliDecoder as well as reading your startup configuration.

One bug, five products, and counting

One bug, five products, and counting — a painterly illustration conveying this idea from "Microsoft Patched .NET for a Bug That Isn't .NET's. Most Apps Were Never Exposed."

The reason this keeps resurfacing is worth understanding, because it will happen again with something else.

Brotli is maintained by Google as a single piece of C code. Rather than everybody calling one shared copy on your machine, projects copy that decoder into their own code and ship it as part of themselves. Node.js has a copy. .NET has a copy. So do a lot of other things. When a flaw is found in the original, every project carrying a copy needs its own patch, released on its own schedule, with its own version number.

That is exactly what has been happening since 21 October. The security firm JFrog, which has been tracking where the decoder turns up, has documented it reaching well past web servers:

Where it surfacedWhat was demonstrated
Node.jsCode execution on an official build (Imperva, 22 Oct)
.NETA native crash; patched out of band 25 Oct
ClickHouseRemote denial of service from a single request
FFmpegInstruction-pointer control in a proof of concept, reached through JPEG XL images
Apache ArrowA 534-byte input causing an infinite loop

Look at the last two. FFmpeg gets there through an image file and Arrow through a data format — no web request, no HTTP, no server. Any mental model of this as “a web thing” is already wrong.

Our Node.js piece, published on 24 October, ended that section with a line that has aged well: “if you run other software that bundles its own Brotli, Node is not the only thing on your list.” It took four days for Microsoft to prove the point.

⚠️ And JFrog adds a caution that matters if you are checking dependencies: “Downstream projects may backport the fix, so the upstream version number alone is not always sufficient to determine patch status.” A library reporting an old Brotli version may well be patched. Trust the project’s advisory over the version string.

The version numbers

Four lines were patched. If you run .NET, you want at least:

.NET linePatched runtime
.NET 88.0.33
.NET 99.0.22
.NET 1010.0.14
.NET 11Release Candidate 3

Check what you have with:

dotnet --list-runtimes

The upstream fix shipped in Brotli 1.2.1, 1.1.1 and 1.0.10, which is the list to check against for anything that is not .NET.

🔴 The same release carries three other fixes, and two of them are re-releases. Microsoft shipped the October 13 updates and then had to go again, saying the earlier release “did not fully address these issues” for CVE-2026-70197 (a memory-corruption bug in socket handling) and CVE-2026-102870 (the Blazor WebAssembly debugger). There is also a separate Linux issue, CVE-2026-114148, where unsafe ownership information in .NET tarballs could let a local attacker modify files and gain privileges — and on .NET 8 and 9 that reaches people installing ASP.NET Core from tarballs, not only those building from source.

If you patched on 13 October and ticked this off, you are not done. That is an unusually easy way to be wrong, and it is nobody’s fault.

Two ways to patch and still be vulnerable

Two ways to patch and still be vulnerable — a painterly illustration conveying this idea from "Microsoft Patched .NET for a Bug That Isn't .NET's. Most Apps Were Never Exposed."

The first is the one Microsoft calls out in bold, and it catches people every time. Updating the .NET runtime on a machine fixes applications that use that shared runtime. It does nothing for an application that brought its own.

The advisory’s wording: “Updating the machine’s shared runtime alone does not replace a decoder already linked into a NativeAOT executable.” The blog says the same thing more generally: “Installing a newer runtime on the host does not update the runtime bundled with your application.”

So if your app is published self-contained, compiled with NativeAOT, or shipped in a container image that carries its own runtime, the fix arrives only when you rebuild and redeploy. Patching the host and watching the version number change on the server is a convincing-looking way to achieve nothing. This is the same mistake as editing a Dockerfile without rebuilding, and it is the single most common way a team believes it has patched and has not.

The second is simpler: restart the application. A long-running service that was already in memory when you installed the runtime keeps running the old one.

If you genuinely cannot deploy today, Microsoft’s interim advice is to stop decompressing untrusted Brotli: remove the br provider from request decompression and keep gzip and deflate, or drop Brotli from your HttpClient handler flags. The advisory attaches a warning to its own workaround — “Do not use these code snippets if you are not already using decompression” — because applying a fix for a feature you never enabled is a good way to break something that was fine. And it is explicit that this is a stopgap: “Installing the security update remains the recommended fix.”

The thing nobody is telling you: .NET 8 and 9 stop getting fixes on 10 November

The thing nobody is telling you: .NET 8 and 9 stop getting fixes on 10 November — a painterly illustration conveying this idea from "Microsoft Patched .NET for a Bug That Isn't .NET's. Most Apps Were Never Exposed."

This is the most practically useful fact in the whole story, and it has been almost entirely crowded out by the CVE.

Per Microsoft’s own support policy, .NET 8 and .NET 9 both reach end of support on 10 November 2026 — fifteen days after this emergency patch shipped. .NET 10 is supported until November 2028.

Let that land. Two of the four runtimes patched here stop receiving security fixes this month. Microsoft’s policy page is blunt about what that means: end of life “refers to the date when Microsoft no longer provides fixes, updates, or online technical assistance,” and “your use of out-of-support .NET versions may put your applications, application data, and computing environment at risk.”

So if you are on .NET 8 or 9, installing 8.0.33 or 9.0.22 is correct and you should do it — but it is the smaller half of the job. The larger half is that the next advisory like this one will not come with a patch for you. That is the same trap we found on the Node.js side, where anyone on Node 20 had no patched version to move to at all, because the line was finished.

Upgrading a major runtime version is an afternoon or a month, not a command. The useful thing about reading this on 26 October rather than 11 November is that you still get to choose when.

Four names for one bug, and one that does not exist

Four names for one bug, and one that does not exist — a painterly illustration conveying this idea from "Microsoft Patched .NET for a Bug That Isn't .NET's. Most Apps Were Never Exposed."

The naming around this has collapsed entirely, and it is worth a minute because it will cost somebody a wasted afternoon.

NameWhere it came from
Node2ShellThe security firm Imperva’s name for its own analysis, 22 October
HTTP2ShellCoined 23 October “to reflect its broader impact on HTTP clients and servers”
Brotli2ShellUsed by JFrog and others from 25 October
Broccoli2ShellCirculating, with its own dedicated website

Not one of those names appears in Microsoft’s advisory, in Node.js’s advisory, or in the CVE record. They are coined by firms writing about the flaw, which is normal and not sinister — but it means a nickname tells you who wrote the article, not what the bug is.

Two of them are worse than useless. HTTP2Shell collides with HTTP/2, a real and completely unrelated internet protocol, and it is wrong on the facts besides: FFmpeg and Apache Arrow reach this bug through image files and data formats with no HTTP anywhere. And one Italian security site had to publish a piece explaining that “html2shell does not exist” — a fifth name, for the same bug, invented by someone who misread one of the other four.

⭐ When we wrote about php2shell, wp2shell and node2shell on 21 October, the problem was three similar names for three different products. This is the opposite and worse: four names for one bug, because the thing that needs a name is not a product at all. It is a dependency that dozens of products copied.

The only identifier that works is the CVE number. Search for CVE-2026-111497. Everything else is marketing.

Who found it

Who found it — a painterly illustration conveying this idea from "Microsoft Patched .NET for a Bug That Isn't .NET's. Most Apps Were Never Exposed."

Credit is the one thing the Reddit and forum versions of this story get consistently wrong, and the correct answer is more interesting anyway.

Node.js’s advisory — the primary record, because this surfaced there first — says, verbatim:

“Thank you, to OpenAI Codex Security – Khai Tran and Sylvie Mayer for reporting this vulnerability and thank you mcollina and aduh95 for fixing it.”

So: reported by two named researchers on OpenAI’s Codex Security team, and fixed by two Node.js maintainers. Imperva, widely credited online with finding it, explicitly does not claim that — it describes its own work as “an independent validation of the impact and the released patch.” Microsoft’s .NET advisory names nobody at all.

It is tempting to file this as “an AI found a vulnerability,” and worth resisting. The advisory credits people, on a team at an AI company. It says nothing about what tools they used, and silence is not evidence either way. What can be said is narrower and still notable: a security team at an AI lab reported a memory-corruption flaw in a compression library that has been in production for a decade, and nobody disputes it.

That lands in a week when an AI model also solved a twenty-nine-year-old mystery about the NIST elliptic curves — a different kind of result, with a different kind of evidence behind it. Two stories, four days apart, both worth reading for what they actually claim rather than what the headline rounds them up to.

How dangerous is it, really?

The honest answer on 26 October 2026 is: serious, not yet urgent, and genuinely contested.

Microsoft’s own assessment is the strongest single data point, and it is more measured than the coverage. It scores the flaw 8.1, High — not Critical — and the technical rating it published includes a metric stating that exploit code maturity is unproven. It also rates the attack complexity as high.

Google, which assigned the CVE, scores it 9.2 Critical — on a newer version of the scoring scale. Node.js published no number at all and called it High. These are not contradictions so much as different instruments; it is normal, and it is why a single scary number quoted without its source is close to meaningless. One outlet reported a score of 10, which does not appear to come from any of the three authorities.

On working exploits, be careful in both directions. You will read that proof-of-concept code is already circulating on GitHub. Imperva’s own write-up describes reaching code execution against the official Node.js 24.22.0 Linux arm64 image, stating that “it needed no debugger or instrumented binary” — but also that “execution took multiple attempts” and “the proof did not succeed on every attempt.” Imperva did not release its exploit. JFrog, which got instruction-pointer control in FFmpeg, notes that “achieving remote code execution would require further exploitation of the memory corruption.”

We got this one wrong ourselves and are fixing it. Our 24 October Node.js article said the public proof of concept worked only against a deliberately weakened build, following a well-regarded outlet’s description of Imperva’s research. Imperva’s own post says otherwise, and Imperva is the authority on what Imperva did. The lesson is not that summaries are untrustworthy — it is that when a claim is about what a specific company did, that company’s own write-up is the document to read, however reputable the publication summarising it.

As of 26 October 2026, this flaw was not listed in CISA’s Known Exploited Vulnerabilities catalogue. One threat-intelligence firm, ELLIO, has reported seeing an exploitation attempt; that report is recent, we could not corroborate it, and it sits alongside same-week statements that no in-the-wild exploitation had been observed. Treat it as a reason to patch on schedule, not as evidence the house is on fire.

Frequently asked questions

I run a WordPress site. Does any of this affect me? Almost certainly not. WordPress runs on PHP, not .NET. If you got here from a headline mentioning php2shell or wp2shell, those are different flaws in different software — the PHP one from the same week is the one to read.

How do I know if Brotli decompression is turned on in my app? Check three places: whether request decompression is registered with the br provider in your ASP.NET Core startup; whether any HttpClientHandler or SocketsHttpHandler sets AutomaticDecompression to include Brotli or All; and whether anything in your code uses BrotliStream or BrotliDecoder directly. The third needs a code search — configuration will not show it.

I installed the update on the server. Am I done? Only if your applications use the machine’s shared runtime, and only after you restart them. Self-contained, NativeAOT and container deployments carry their own runtime and need a rebuild and redeploy.

I already patched on 13 October. Then you still need this one. Microsoft re-released two of the October advisories because that release “did not fully address these issues,” on top of the Brotli flaw, which is new.

Should I block Brotli at my firewall? That is not the recommendation, and it is a weaker control than it sounds — a firewall is poorly placed to judge whether a compressed payload will misbehave once expanded. Microsoft’s interim advice is to stop your own application from decompressing Brotli, which is a different and more reliable thing. Then patch.

Is .NET Framework affected? No. Microsoft states explicitly that .NET Framework is not affected.

I am on .NET 8. Is that still supported? Until 10 November 2026, and then no. Install 8.0.33 now, and start planning the move to .NET 10, which is supported into 2028.

✅ Pro tip: when a flaw lands in a library rather than a product, the useful question changes from “is there a patch?” to “what else did I ship that contains a copy of this?” The answer for Brotli, in five days, was a JavaScript runtime, a Microsoft runtime, a database, a media framework and a data library — and that list is almost certainly not finished. Write down what your software is actually built out of while nothing is on fire. On the day of the advisory, the teams that patched quickly were not better informed than everyone else. They just already had the list.



📦 Recommended Resources

Some links below are Amazon affiliate links. If you buy through them we earn a small commission at no extra cost to you.

Read the free primary material first, because in this case it is short and unusually clear. Microsoft’s advisory for CVE-2026-111497 is the authoritative list of affected and patched versions, and it does something most advisories do not: it tells you plainly when you are not affected. The .NET engineering blog post covers the other three fixes in the same release, including the two that were re-issued because October’s did not land. And Microsoft’s .NET support policy page is the one to bookmark — it is where you find out that .NET 8 and 9 stop receiving fixes on 10 November 2026, which will matter longer than this CVE does. All three were read on 26 October 2026.

The three books below are for the part that is still true next year, when this release’s version numbers mean nothing to anybody.

  • Software Engineering at Google — Titus Winters, Tom Manshreck & Hyrum Wright · Book. The right book for the actual lesson here, which is not about Brotli. Its central argument is that software engineering is programming integrated over time, and its chapters on dependency management deal directly with the situation this week produced: code you did not write, copied into things you did not build, that becomes your problem on somebody else’s schedule. If you have ever wondered why “just upgrade” is a sentence that makes experienced engineers wince, this is the explanation.
  • Building Secure and Reliable Systems — Google SRE authors · Book. The practical treatment of the capability this week tested, which was never knowledge and always deployment. Its chapters on keeping systems updatable and designing for recovery explain why some teams shipped 8.0.33 in an afternoon while others are still scheduling it — they had already answered “what am I running, and how fast can I replace it?” before anyone needed the answer.
  • Security Engineering: A Guide to Building Dependable Distributed Systems — Ross Anderson · Book. Read this for the bug itself. An out-of-bounds write triggered by malformed compressed input is a textbook example of the class Anderson is best on: nothing is broken in isolation, every component is doing roughly what it was asked, and the vulnerability lives in the handling of input nobody expected to receive. It is also the best single explanation of why memory-safety flaws in decoders keep being the ones that matter.

Newsletter

Tags

Agentic AI Agentic Development AI AI Agents AI Coding Agents Blockchain Bounded Context Browser ChatGPT Plugins Claude Code CLPR Context Engineering Cryptocurrency DePin Developer Tools Distributed Systems DLT Domain-Driven Design Google Search HBAR Hedera Home Networking Internet Internet Performance LLM Machine Learning MCP Model Context Protocol Network Architecture Network Diagnostics Network Security News NLP Postgres PostgreSQL Productivity RAG Routers Saga Pattern SEO Software Architecture software engineering Troubleshooting Twitter Web3

Recent Articles

  • AI Agent Memory: Search and RAG Limits November 6, 2026
  • Plan Locally, Execute in the Cloud: Claude Code’s Own Recommended Workflow November 6, 2026
  • The AI Evaluation Scorer That Never Read a Transcript November 6, 2026
  • What Is a Browser Engine? Blink & V8 Explained November 5, 2026
  • Why Postgres Is Winning: The $1B Acquisition Wave November 5, 2026
  • RENDER Tokenomics Explained: Burn, Mint, Repeat November 5, 2026

Copyright © 2026 · News Pro Theme On Genesis Framework · WordPress · Log in

Content on this site is produced with the assistance of Large Language Models and edited by a human. It is intended for educational purposes — facts may occasionally be incorrect. If you spot an error, please email Larionovdmitri@learninternetgrow.com.

Affiliate Disclosure: This site contains affiliate links to Amazon and other partners. If you purchase through these links, LearnInternetGrow.com may earn a commission at no additional cost to you. We only recommend products we genuinely believe in.

Privacy Policy · Terms of Service