Plex published a security note for Plex Media Server 1.43.3 on 1 September with no CVE numbers in it. The details arrived on 19 October, once the identifiers existed and exploitation had started 1. All four sit in one product, an attacker chains them into code execution, and CISA added the set to the KEV catalog on 20 October, with 23 October as the due date for CVE-2026-96654 and CVE-2026-96656 and 3 November for the two request forgery bugs 2. A public proof of concept appeared on 16 October, and Plex reports widespread attacks against internet-facing servers that have not updated since 1.
The chain in the order it runs
Each stage looks harmless when it stands alone.
- CVE-2026-96652, the way in. The
/player/timelineendpoint checks that anX-Plex-Tokenheader exists, not that the value means anything, so a junk string passes. Theprotocolparameter then goes into a callback URL without escaping, and the server POSTs to whatever address the attacker put there 3. - CVE-2026-96654, the pivot. That POST arrives from loopback, and the plugin API under
/system/agentsanswers loopback requests without a token. The identifier guard insearchOnediscards its own return value, so an attacker forges a call to any plugin function 3. - CVE-2026-96655, the token. The forged call reaches the plugin framework’s
GetResultFromNetworkwith a URL likehttp://127.0.0.1.<attacker>.sslip.io. The framework adds the owner’s admin token to any outbound request whose URL starts withhttp://127.0.0.1, and that test is a string prefix, so the token travels to an address the attacker controls 3. Plex calls this flaw disclosure of the server owner’s admin token, and the CVE record for the same identifier describes a request forgery through the transcode path parameter 14. - CVE-2026-96656, the payload. Holding the owner token, an attacker writes to
/:/prefs, setsTranscoderH264Optionstodump-yuv, and the transcoder writes decoded frames into the Codecs directory as a shared object. The next transcode loads it, and the constructor runs as the Plex user 3.
The CVE records carry 5.3 for both request forgery stages, and NVD scores the file write 8.6 and the plugin call 6.9, with every score lower under CVSS 3.1 456. Plex states the chain as a whole lets an unauthenticated attacker run code with the privileges of the Plex process 1.
Why I patch this as one unit
The two request forgery bugs sit at the bottom of the score list and they are the ones that make everything above them usable. A blind POST from a server to itself reads like housekeeping. The 8.6 needs the highest privilege the product has. Chained, the low-scored stage hands over the token that satisfies the high-scored one.
CISA’s deadlines split along the same lines, and that split is the trap. Sorting the four by due date puts the file write first and leaves a live path to the admin token open until 3 November. I work the order backwards from the score, because one version bump closes the chain and 1.43.3.10861 is the release that contains all four fixes 1.
What a request forgery reaches from my media network
My server makes the request, from an address and a network namespace I gave it. Whatever the container can route to, an attacker can POST to, blind, with nothing coming back.
So the work started with the reachable set. Four questions decide it:
- The container’s DNS resolver. If it answers internal names, an attacker gains a list of services to walk through by name.
- The networks the container shares. My media stack sits on its own overlay, so the answer is the reverse proxy and little else. A bridged container inherits far more of the LAN.
- The host gateway and any management plane. That answer lives in the network config.
- The files and tokens the Plex service account can read. Mounted shares, credentials in its own config, anything else on the same machine.
On my stack the reachable set came back short: the reverse proxy, the resolver, the internet. The first two are there by design, and both accept requests from inside the media network that they refuse from outside it, so a forged request opens a conversation the attacker cannot open directly.
A URL prefix check fails as an allowlist once wildcard DNS is in play, because 127.0.0.1.anything.sslip.io passes the string check and then resolves wherever the attacker points it 3. The transcode fetch has the same shape: no host allowlist, so an attacker aims the server at a file they host and the server fetches it 3.
The payload is the smaller problem
Plex reports compromised servers mining cryptocurrency and running proxy software that routes other people’s traffic through the victim’s connection 1. The miner costs electricity. The proxy spends the reputation of an IP address that now carries someone else’s traffic, and the abuse report goes to whoever owns the connection.
Plex lists what to look for: processes under Plex named myawesomeproxy, mylovelyminer or x12345, unfamiliar .so or .dll files in the Codecs folder, outbound connections to proxystream.st, and sustained CPU use with nothing playing 1.
The order I work in
Version first, because every other check depends on what version answered. On Docker, pulling a new image is not the same as running it. My tags are pinned, so an update means a tag bump and a container recreate, which puts the version change in front of me where I can see it.
Then the compromise check, before I touch exposure. If a server on my stack showed any of the indicators and had been reachable from outside, I would rebuild it rather than clean it, leave the old data directory behind, sign out devices from the Plex account and change the password. Plex recommends treating whatever the service account could reach as exposed 1, and on a home stack that list runs longer than the owner expects.
Exposure comes third. Remote access off if nobody uses it, no port forward straight to 32400, and the reverse proxy in front of anything that stays reachable.
Mine runs 1.43.3.10861, the Codecs directory holds only what Plex downloaded itself, and 32400 stays closed from the internet.
-
Plex. (2026, October 19). Important security update for Plex Media Server v1.43.2 and earlier [Forum announcement]. Plex Forums. https://forums.plex.tv/t/important-security-update-for-plex-media-server-v1-43-2-and-earlier/942319 ↩︎ ↩︎ ↩︎ ↩︎ ↩︎ ↩︎ ↩︎ ↩︎
-
Cybersecurity and Infrastructure Security Agency. (2026, October 20). Known exploited vulnerabilities catalog [Data set]. https://www.cisa.gov/sites/default/files/feeds/known_exploited_vulnerabilities.json ↩︎
-
Main, Z. (2026, October 16). Plex2Shell: Unauthenticated RCE in Plex Media Server. https://zmain.info/blog/plex2shell ↩︎ ↩︎ ↩︎ ↩︎ ↩︎ ↩︎
-
Cybersecurity and Infrastructure Security Agency. (2026). CVE-2026-96655 record. CVE Program. https://www.cve.org/CVERecord?id=CVE-2026-96655 ↩︎ ↩︎
-
National Institute of Standards and Technology. (2026). CVE-2026-96656 detail. National Vulnerability Database. https://nvd.nist.gov/vuln/detail/CVE-2026-96656 ↩︎
-
National Institute of Standards and Technology. (2026). CVE-2026-96654 detail. National Vulnerability Database. https://nvd.nist.gov/vuln/detail/CVE-2026-96654 ↩︎