Guide · Ilia Dubovskii · 17 November 2026

The *2Shell vulnerabilities: React2Shell, PHP2Shell, Node2Shell, Java2Shell, and a playbook for the next one

Since December 2025 a run of critical, unauthenticated remote code execution bugs has landed in the runtimes and frameworks almost everyone uses, each nicknamed "something2Shell": React2Shell, PHP2Shell, Node2Shell (with its aliases HTTP2Shell and Python2Shell) and now Java2Shell. This page lists them all with CVEs and status, says which names people search for that aren't real yet, and sets out a playbook for whichever one comes next.

The short version

1. Every *2Shell and *4Shell, in one table

Name CVE Where the bug is Disclosed Status (17 Nov 2026)
Log4ShellCVE-2021-44228Log4j 2 library: JNDI lookups in logged stringsDec 2021Fixed; mass-exploited
Spring4ShellCVE-2022-22965Spring Framework data binding on JDK 9+Mar 2022Fixed; exploited
Text4ShellCVE-2022-42889Apache Commons Text string interpolationOct 2022Fixed; narrow real-world exposure
React2ShellCVE-2025-55182 (CVE-2025-66478 rejected as a duplicate)React Server Components; Next.js App Router3 Dec 2025Fixed; exploited within days; in KEV
PHP2ShellCVE-2026-109618, CVE-2026-109619The PHP interpreter, every web SAPI20–22 Oct 2026Fixed 21 Oct; exploited before the fix; CVE-2026-109618 in KEV
Node2Shell / HTTP2Shell / Broccoli2ShellCVE-2026-111497Google's Brotli decoder in Node.js, Bun, .NET, Envoy, Istio21–22 Oct 2026Fixed; exploitation reports disagree
Py2Shell / Python2ShellCVE-2026-111497The same decoder in PyPI's Brotli and brotlicffi, used by requests, urllib3, httpx, aiohttp23 Oct 2026Fixed in Brotli 1.2.1, brotlicffi 1.2.1.0
Java2Shell / Java4ShellCVE-2026-114938Reported: OpenJDK 17.0.6+ (memory corruption)16–17 Nov 2026Reported, unconfirmed; no fix confirmed; no exploitation reported

2. Names without a CVE (yet)

After Node2Shell, people online joked about the rest of the series: Ruby2Shell, Go2Shell, Dotnet2Shell, Rust2Shell. As of 17 November 2026, none of those is a real vulnerability name. If you got here searching for one, here's what's actually relevant:

3. What they have in common

4. A playbook for the next one

You can't know which runtime is next. You can make the next one boring. Five parts, roughly in the order they pay off:

  1. An inventory you can query. "Which internet-facing services run PHP 8.3 below .36?" should take minutes. That means knowing the image each running task uses, what each Dockerfile installs, and what each lockfile pins, linked to whether the workload faces the internet. Disclosure day is the wrong day to build that.
  2. A patch path that's a pull request, not a project. One pull request per repository with the version bumped, CI and the image build checked, and a rollback written down. A person merges. Critical, exploited, internet-facing findings shouldn't wait for someone to open a ticket.
  3. A tripwire for the window. There will be days between exploitation and the patch reaching production. A honeytoken, a decoy credential in each workload's environment that nothing legitimate uses, turns the attacker's predictable first move into an alert with almost no false positives, and tells you which workload it came from.
  4. One automatic, undoable response. Because a honeytoken trip is near-certain, it's safe to act on without a human: restart the container from its image, or deactivate the key. Both can be undone. That's the principle we call automate on certainty.
  5. Alerts as context, not triggers. Keep GuardDuty, Falco and your CPU and egress alerts. When a honeytoken trips, they explain what the attacker did. Automating on them directly means acting on symptoms that usually have innocent causes.

With that in place, our PHP2Shell incident was contained in under two minutes, ten months after React2Shell took four days.

5. Where Compass fits

Compass is our open-source (Apache 2.0) operations agent for AWS, and it implements four of the five parts:

The fifth part, the alerts, you already have. Compass doesn't replace your WAF, GuardDuty or runtime monitoring, and doesn't stop the exploit request itself. More about Compass.

6. FAQ

What does "2Shell" mean in a vulnerability name?

It's shorthand for "from X to a shell": a bug that takes an attacker from sending a request to running commands on the server. The naming started with Log4Shell in 2021, then Spring4Shell and Text4Shell, and from React2Shell in December 2025 the "2Shell" form took over. It isn't an official classification; researchers and the press pick the nickname.

Is there a Ruby2Shell, Go2Shell or Dotnet2Shell vulnerability?

Not as of 17 November 2026: no CVE carries those nicknames. People joked about the series continuing after Node2Shell. However, the Ruby Brotli gem and .NET 8, 9 and 10 were affected by CVE-2026-111497, the bug called Node2Shell and HTTP2Shell, so if you run Ruby or .NET with Brotli decompression, check those fixes.

Are Node2Shell, HTTP2Shell, Python2Shell and Broccoli2Shell different bugs?

No. They're all CVE-2026-111497, an out-of-bounds write in Google's Brotli decoder, which Node.js, Python's Brotli packages, Bun, .NET, Envoy and Istio all ship. The name shifted as more runtimes were found to be affected.

Are Java2Shell and Java4Shell the same?

Yes, both names refer to CVE-2026-114938, a reported unauthenticated remote code execution in OpenJDK 17.0.6 and later. As of 17 November 2026 the details were not confirmed by Oracle or the OpenJDK project.

Which *2Shell vulnerabilities are in CISA's KEV catalogue?

React2Shell (CVE-2025-55182) and PHP2Shell's CVE-2026-109618 are in the Known Exploited Vulnerabilities catalogue. Reports disagree on whether CVE-2026-111497 (Node2Shell) was added. Java2Shell was not listed as of 17 November 2026. Check the catalogue directly, because it changes daily.

How do I prepare for the next *2Shell vulnerability?

Know what runs where, so you can answer "are we affected?" in minutes. Make patching a pull request per repository, not a project. Put a honeytoken (a decoy credential) in each workload's environment, because attackers read the environment first. Automate one undoable response on a honeytoken trip, such as restarting the container. And keep your alerts as context rather than triggers.

Written by Ilia Dubovskii, founder of Beamreach. CVEs, dates and statuses were checked against public advisories and security-vendor reporting on 17 November 2026. Java2Shell is still developing, and reports on Node2Shell's exploitation disagree; each page has the details and sources.

Related: React2Shell · PHP2Shell · Node2Shell · Py2Shell · Java2Shell · What to automate in incident response · One Fake Password Is Worth More Than 10,000 Alerts (talk)