P.K. SHARMA

Cyber security intelligence, AI governance, practitioner analysis

NGINX 1.31.9 fixes ten flaws; F5 scores five 8.1 and 9.2, nginx.org rates three medium; HTTP/2 range from 2015

nginx.org lists ten fixes in NGINX 1.31.9 and 1.30.7, and F5's CVE records score five of them 8.1 and 9.2; nginx.org's own word for three of those five is medium. The HTTP/2 flaw's printed range starts at 1.9.5 in September 2015, and CISA's entries say exploitation none.

By · LinkedIn · · 22 min read

Ten fixes, five scored 8.1 and 9.2, and a project word that says medium for three of them

nginx.org's advisory page lists ten entries that read "Not vulnerable: 1.31.9+" and, for nine of them, "1.30.7+" as well (derived count: ten; the PROXY protocol version 2 crash lists 1.31.9 only). The CVE records F5 published for all ten at 15:23 UTC on 18 November 2026 give five of them a CVSS 3.1 score of 8.1 and a CVSS 4.0 score of 9.2. nginx.org's own word for three of those five is medium (the resolver, HTTP/3 and HTTP/2 use-after-free flaws), and for the other two, HTTP/2 flow control and the mail proxy, it is major. This site reports both ratings and picks neither. The same split between project and vendor appeared in the 14 October release, where nginx.org rated all five of that release medium and F5's CVSS 4.0 put three at 9.2.

The HTTP/2 flaw, CVE-2026-123007, shows the split in one place. nginx.org's page says "Severity: major" and lists vulnerable versions 1.9.5 to 1.31.8. The project's change log says a flow control window overflow "might have potential other impact". F5's CVE record scores it 8.1 and 9.2 and says "attackers can execute code on systems with Address Space Layout Randomization (ASLR) disabled or when the attacker can bypass ASLR". 1.9.5 is dated 22 September 2015 in the change log and 1.31.9 is dated 18 November 2026: that is 4,075 days, or 11 years 1 month 27 days (derived). So a headline that calls the flaw ten years old is under by the page's own dates; the relay that carried such a headline was not read, so this site says only that the saved dates give more than 11 years.

What that does not establish. No saved page says anyone has exploited any of the ten: the CISA-ADP entry on each record says exploitation none. No page says what "undisclosed traffic", the record's phrase for the HTTP/2 trigger, is. No page says how many servers run an affected build, whether distributions have shipped fixed packages, or how common it is for ASLR to be disabled or bypassed on the hosts that run NGINX. This site does not claim the flaw can be exploited on a default build with ASLR on, and no saved page claims it either. Every figure this site worked out is labelled "derived".

Update, 19 November 2026, about 12:40 UTC. The pages this site had saved when this brief was published did not include the researchers' own write ups. So the sentence above that no saved page claims code execution with ASLR on, and the table row that says there is no proof of concept, describe the nginx.org and F5 pages only. Later pages say more: Rapid7 says it built a proof of concept for CVE-2026-123007 and that code execution needs about 128 GB of HTTP/2 data, Wiz says it ran code on NGINX serving ordinary static files and that ASLR does not protect, and Censys says it reproduced the accounting error with a small test. A later brief sets out what each of those pages says and what none of them establishes. Nothing above about nginx.org's words, F5's scores or CISA's entries has changed.

Whose words these are. NGINX is F5's product: nginx.org is the project's site, F5 is the assigner of the CVE records and the commercial vendor of NGINX Plus, so a severity word and a score both come from parties with a stake in how urgent the release reads. This site names no individual: the change log and the records credit people, and only organisations are named here.

The ten entries side by side

The table gives, for each entry on the nginx.org page for the 18 November release, the project's title and word, the vulnerable range it prints, and the two scores and the CISA-ADP SSVC entry from the F5 record. SSVC (Stakeholder-Specific Vulnerability Categorization) is the decision scheme CISA's entries use; each entry prints three options: Exploitation, Automatable and Technical Impact. nginx.org prints one id as "CVE-CVE-2026-121810"; the record's id is CVE-2026-121810, and the table uses the record's.

The ten entries for the 18 November 2026 release as read on 19 November 2026: nginx.org's advisory page (title, severity word, vulnerable range) and F5's CVE records (CVSS 3.1 then 4.0 base score and the record's severity word; CISA-ADP SSVC options in the order Exploitation, Automatable, Technical Impact).

Flaw as titled by nginx.orgnginx.org severity and rangeF5 scores and CISA SSVC
Use-after-free during response processing in resolver (CVE-2026-123634)medium; vulnerable 0.6.18-1.31.88.1 HIGH / 9.2 CRITICAL; SSVC none, no, total
Memory corruption in HTTP/2 flow control (CVE-2026-123007)major; vulnerable 1.9.5-1.31.88.1 HIGH / 9.2 CRITICAL; SSVC none, no, total
Use-after-free in HTTP/3 (CVE-2026-121810)medium; vulnerable 1.25.0-1.31.88.1 HIGH / 9.2 CRITICAL; SSVC none, no, total
Use-after-free in HTTP/2 (CVE-2026-123633)medium; vulnerable 1.19.7-1.31.88.1 HIGH / 9.2 CRITICAL; SSVC none, no, total
Infinite loop while reading client request body over TLS 1.3 (CVE-2026-118899)major; vulnerable 1.13.0-1.31.87.5 HIGH / 8.7 HIGH; SSVC none, yes, partial
Use-after-free in mail proxy module (CVE-2026-113425)major; vulnerable 0.5.33-1.31.88.1 HIGH / 9.2 CRITICAL; SSVC none, no, total
NULL pointer dereference when using PROXY protocol version 2 (CVE-2026-120993)minor; vulnerable 1.31.4-1.31.85.3 MEDIUM / 6.9 MEDIUM; SSVC none, yes, partial
Buffer overflow when using alias with try_files (CVE-2026-125283)minor; vulnerable 0.9.6-1.31.85.3 MEDIUM / 6.9 MEDIUM; SSVC none, yes, partial
SNI bypass when using client certificate validation in stream (CVE-2026-120967)medium; vulnerable 1.25.5-1.31.84.8 MEDIUM / 6.3 MEDIUM; SSVC none, no, partial
Insufficient validation in stream ssl_preread module (CVE-2026-123640)minor; vulnerable 1.15.2-1.31.85.3 MEDIUM / 6.9 MEDIUM; SSVC none, yes, partial
Drawn from the nginx.org advisory page and the ten F5 CVE records, read on 19 November 2026. Bar lengths are to scale. The row order and the counts are derived.

Counted from the page and the records (derived): the severity words are major 3, medium 4 and minor 3. The CVSS pairs are 8.1 and 9.2 on five records, 7.5 and 8.7 on one (the TLS 1.3 loop), 5.3 and 6.9 on three, and 4.8 and 6.3 on one (the stream client certificate bypass). Read in plain words, nginx.org's word is lower than the record's CVSS 4.0 word on six of the ten: medium against CRITICAL on three and minor against MEDIUM on three. The two major rows at 9.2 and the major TLS loop at 8.7 (HIGH) are not in that group. SSVC says exploitation none on all ten; technical impact total on exactly the five 9.2 records and partial on the other five; automatable yes on four records, the TLS loop and the three scored 6.9, and no on the six others. None of the four automatable records is a 9.2 record, so the three vocabularies do not rank the ten in the same order, and the records do not say how to combine them.

Three vocabularies for the HTTP/2 flow control flaw

For CVE-2026-123007 three saved pages use three vocabularies, set side by side below. The table copies short phrases exactly; the sources are named in words.

How three saved pages describe the HTTP/2 flow control flaw, as read on 19 November 2026. Quotations are short and exact.

PageWhat it printsWhat it states
nginx.org advisory page"Memory corruption in HTTP/2 flow control"; "Severity: major"; Vulnerable 1.9.5-1.31.8A project word and a range. No definition of its words is printed on the page.
nginx.org change log, 1.31.9, 18 Nov 2026"a flow control window overflow in HTTP/2 might allow an attacker to cause a segmentation fault in a worker process, or might have potential other impact"A crash is named; anything beyond it is "potential other impact", not described.
F5 CVE record, description"potentially resulting in a heap buffer overflow typically followed by an NGINX worker segmentation fault"A heap buffer overflow, usually ending in a worker crash.
F5 CVE record, description"attackers can execute code on systems with Address Space Layout Randomization (ASLR) disabled or when the attacker can bypass ASLR"Code execution where ASLR is off or can be bypassed.
F5 CVE record, impact"crash the NGINX worker process, corrupt its memory, or potentially execute code remotely"Crash, memory corruption or potential remote code execution, all conditional.
F5 CVE record, scoresCVSS 3.1 8.1 HIGH and CVSS 4.0 9.2 CRITICALThe same base metrics under two versions of the scoring system (derived from the two vector strings).
CISA-ADP entry in the recordExploitation: none; Automatable: no; Technical Impact: totalA coordinator's reading at the time stamp the record prints.

The three are not contradictory word for word: each hedges. "Might", "potentially" and "could" run through all three, and none says code execution has been done. But they sit at different distances from the reader's decision. "Major" is a rank on a scale the project's page does not define. "Might have potential other impact" names no impact, and the change log does not describe one. 9.2 CRITICAL is a calculation from a vector that rates attack complexity High and the impact on confidentiality, integrity and availability all High (derived from the vector string). The 8.1 and 9.2 are one vector read by two versions: this site made the same point for a Java flaw in an earlier briefing.

The same "code execution where ASLR is disabled or bypassed" sentence appears in four of the five 9.2 records in near identical words and in the fifth, the resolver record, in slightly different words (derived from the records). ASLR, address space layout randomisation, is a host and kernel setting, not an NGINX setting. The records do not say how often it is off, or what bypassing it would take.

What each of the five 9.2 records says it needs is where the readers' decision actually sits. The table puts each record's own condition beside the change log's wording.

Conditions for the five records scored 8.1 and 9.2, from the F5 records and the nginx.org change log (1.31.9), read on 19 November 2026. Wording shortened; quotations exact.

Flaw (nginx.org word)Condition in the F5 recordWorkaround the record prints
HTTP/2 flow control (major)"configurations that use HTTP/2"; the trigger is "undisclosed traffic"None (a patch is the only remedy the record prints)
HTTP/2 use-after-free (medium)HTTP/2 for client connections, proxying to a backend; "if this process is running low on worker_connections"None
HTTP/3 use-after-free (medium)configurations that use HTTP/3; "Exploitation also depends on conditions beyond the attacker's control"None
Resolver use-after-free (medium)a proxy whose proxied server name includes variables, and an SSI or mirror directive that sends a subrequest to a location resolving the same nameAvoid using SSI or mirror directives and avoid configuring a proxied server name with variables
Mail proxy use-after-free (major)NGINX configured as a mail proxy; "undisclosed traffic"Restrict listeners to trusted networks.

The three nginx.org "medium" records are each tied to a stated condition that looks narrow: the resolver flaw needs a particular proxy and subrequest arrangement; the change log says the HTTP/2 and HTTP/3 use-after-free flaws occur if worker_connections were not enough, the HTTP/2 record repeats that, and the HTTP/3 record says exploitation depends on conditions beyond the attacker's control. A narrow condition would fit a project calling them medium, and a 9.2 is a score that does not weigh it (inference); the records do not reconcile the two. A defender who reads only "medium" or only "9.2" skips the one thing the records print that is specific to a server: the condition.

A range that starts in 2015, and a release cycle of four weeks

The nginx.org page prints the HTTP/2 flaw's vulnerable range as 1.9.5 to 1.31.8. The change log's 1.9.5 entry, dated 22 September 2015, lists "Feature: the ngx_http_v2_module (replaces ngx_http_spdy_module)." So the range starts at the release whose change log entry introduces the HTTP/2 module as a feature (this site's reading of the two pages together). From that date to the 1.31.9 fix on 18 November 2026 is 4,075 days (derived: 22 September 2015 plus 11 years is 22 September 2026, plus 57 days is 18 November 2026), which is 11 years 1 month 27 days (derived).

A printed range is not a discovery date. The pages print version numbers, not the date anyone found the flaw, and they do not say it could have been reached in the same way in 2015. HTTP/2 is also not the oldest range in this release. The resolver flaw's range starts at 0.6.18, dated 27 November 2007 in the change log, 6,931 days before the fix (18 years 11 months 22 days, derived). The alias with try_files range starts at 0.9.6, dated 21 March 2011, 5,721 days before the fix (15 years 7 months 28 days, derived). The mail proxy range starts at 0.5.33, a version whose date the saved change log does not print.

Drawn from the nginx.org advisory page and the nginx.org change log, read on 19 November 2026. Positions are to scale. The 4,075 day span and the other day counts are derived.

The release gap is short. 1.31.8 is dated 21 October 2026 and 1.31.9 is dated 18 November 2026, 28 days apart (derived: four weeks). The saved change log lists ten 1.31.x releases from 1.31.0 on 13 May 2026 to 1.31.9 on 18 November, 189 days (derived), and seven of the ten carry entries on the nginx.org page (counted from each entry's first listed fixed version): 29 entries in all, so the ten of 18 November are 34 per cent of them (derived: 10 of 29). Three of the ten releases, 1.31.4, 1.31.5 and 1.31.8, carry none (derived).

The 1.31.x releases from the nginx.org change log, with the days since the previous release (derived) and the number of advisory entries whose first listed fixed version is that release (derived from the advisory page). The 1.30.x stable releases are not dated in the saved change log.

ReleaseDate in the change logDays since previous (derived)Entries on the advisory page (derived)
1.31.013 May 2026first6
1.31.122 May 202691
1.31.217 Jun 2026263
1.31.315 Jul 2026283
1.31.419 Aug 2026350
1.31.52 Sep 2026140
1.31.615 Sep 2026131
1.31.714 Oct 2026295
1.31.821 Oct 202670
1.31.918 Nov 20262810

What the pages state, and what they do not

The negative column is the point of this table: for each question a reader will ask, what a saved page prints and what none of them does.

What the nginx.org page, the change log and the F5 records state and do not state, read on 19 November 2026.

QuestionStated on a saved pageNot stated on any saved page
Which builds are fixed?Not vulnerable from 1.31.9 and 1.30.7; the PROXY v2 entry lists 1.31.9 onlyAny distribution's fixed package or its date; the date of 1.30.7
Is it exploited?CISA-ADP SSVC: exploitation none, on all ten recordsAnything after those entries; any demonstration of code execution
What triggers the HTTP/2 flaw?"undisclosed traffic" on configurations that use HTTP/2The traffic itself, or any other required setting
When can code run?Where ASLR is disabled or the attacker can bypass it (five records)How common that is, or what a bypass takes
How many servers?NothingThe share of internet facing NGINX on an affected build
Is there a workaround?None printed for the two HTTP/2 flaws, HTTP/3 and TLS 1.3; one for each of the other sixWhether the printed ones are enough in every configuration
Are there indicators or a proof of concept?Nothing: no indicators, no proof of concept, no triggerA log line or signature for the flaws
What do the severity words mean?Major, medium and minor appear; no definition on the pageHow nginx.org chooses a word, or how it relates to CVSS

Where the pages differ, left unresolved

This site reports each difference as printed and resolves none.

The project's word against the vendor's score. Six of the ten rows (see the table above): nginx.org's word is lower than the record's CVSS 4.0 word. Whether the project, the vendor, or neither has the better reading is not something the pages let a reader settle.

"Segmentation fault" against "buffer overflow" for try_files. nginx.org's title and the change log call CVE-2026-125283 a buffer overflow ("a buffer overflow might occur in a worker process"); F5's record says certain capture combinations "can cause a segmentation fault in an NGINX worker process" and files the weakness as an integer underflow, with an availability impact only in its vector (low availability impact, none for confidentiality and integrity). nginx.org's word for it is minor and the record's severity word for the score is MEDIUM.

A typo and a version. nginx.org prints "CVE-CVE-2026-121810"; the record prints "CVE-2026-121810". The HTTP/3 record lists its open source range from "1.25.09", where nginx.org says 1.25.0. Both are as printed.

Dates. The nginx.org advisory page prints no release date. The date of 1.31.9, 18 November 2026, comes from the change log, and the records give the publication time, 15:23 UTC on 18 November. The change log the saved page holds is the mainline log: 1.30.7 is not dated in it.

SSVC entries. The five 9.2 records carry an SSVC time stamp of 00:00:00 on 18 November, before F5 published the records at 15:23 UTC that day; the other five carry time stamps between 16:32 and 16:35 UTC on 18 November. Automatable is yes on four of the five lower scored records and no on the five 9.2 records and the stream client certificate record (derived). The pages print these as they are, and say nothing about why the order differs.

Who found them. The change log's thanks for the HTTP/2 use-after-free names Vexera AI; the record's four credit lines name Wiz Research, Xint.io, Palo Alto Networks Unit 42 and an individual, and not Vexera AI. The change log thanks Xint.io for four of the five 9.2 flaws; the records credit Xint.io on all five (derived). Two of the records' credit lines name an AI tool beside the finder: "Ikotas Labs, Inc." "using GPT-6 Astra" on the HTTP/2 flow control record, and BreachX Zero Day Labs "using Typhon AI Mil v2" on the ssl_preread record. These are F5's credit lines as printed. The records do not say who makes either tool, and this site did not check the claim.

What was added after publication. The HTTP/2 record was updated at 05:39 UTC on 19 November by the CVE Program, which added fifteen references. Among them are pages at Wiz, Rapid7 and Xint.io and a separate site, and the addresses of three of them include the words "rce" or "remote-code-execution". This site did not open them, so it cannot say what they show, and the statement above that no saved page reports exploitation or demonstrated code execution is limited to the pages this site read. A reader should read those pages and the F5 article before deciding how worried to be.

NGINX Plus is in the records; nginx.org's page is about the open source numbers

All ten F5 records list NGINX Plus as affected beside NGINX Open Source. For the HTTP/2 flow control record the printed Plus ranges are 37.1.0 up to below 37.3.0.1, 37.0.0 up to below 37.0.8.1, and release R33 with no upper bound and no fixed version printed. The ranges are not the same in every record: the TLS 1.3 record, for example, lists 37.2.0.1 up to below 37.2.1.1 for the 37.2 line, and the PROXY v2 record lists 37.2.0.1 up to below 37.3.0.1 and R33. Each record adds that "Software versions which have reached End of Technical Support (EoTS) are not evaluated." nginx.org's page speaks only in the open source numbers, 1.31.9 and 1.30.7, and does not mention Plus. A Plus customer therefore needs the record or the F5 article for the release line they run; the pages this site read do not give one table for all.

Builds on branches older than 1.30 fall inside the printed vulnerable ranges, but none is listed as fixed, and the records say end of support versions are not evaluated. This site cannot say from these pages which older branches F5 or the project will patch, only that none is listed as fixed.

What the records print: scores, ranges, workarounds and article ids

These are the artifacts the saved pages print. They are copied from the F5 records and the nginx.org pages, each attributed in its caption, and none of them is a trigger, a payload or an exploit. The pages print no indicators of compromise, no proof of concept, no log line and no detection rule, and this site invents none.

First the vector strings. The five 9.2 records share one pair of vectors; reading them shows the base metrics are the same under both versions (network, attack complexity high, no privileges, no user interaction, high impact on confidentiality, integrity and availability), which is why the two scores move together.

Text
Five records, 9.2 (resolver, HTTP/2 flow control, HTTP/3, HTTP/2, mail proxy):
  CVSS:3.1/AV:N/AC:H/PR:N/UI:N/S:U/C:H/I:H/A:H
  CVSS:4.0/AV:N/AC:H/AT:N/PR:N/UI:N/VC:H/VI:H/VA:H/SC:N/SI:N/SA:N
TLS 1.3 request body loop (CVE-2026-118899), 7.5 and 8.7:
  CVSS:3.1/AV:N/AC:L/PR:N/UI:N/S:U/C:N/I:N/A:H
  CVSS:4.0/AV:N/AC:L/AT:N/PR:N/UI:N/VC:N/VI:N/VA:H/SC:N/SI:N/SA:N
PROXY v2 (CVE-2026-120993), alias with try_files (CVE-2026-125283), 5.3 and 6.9:
  CVSS:3.1/AV:N/AC:L/PR:N/UI:N/S:U/C:N/I:N/A:L
  CVSS:4.0/AV:N/AC:L/AT:N/PR:N/UI:N/VC:N/VI:N/VA:L/SC:N/SI:N/SA:N
ssl_preread (CVE-2026-123640), 5.3 and 6.9:
  CVSS:3.1/AV:N/AC:L/PR:N/UI:N/S:U/C:N/I:L/A:N
  CVSS:4.0/AV:N/AC:L/AT:N/PR:N/UI:N/VC:N/VI:L/VA:N/SC:N/SI:N/SA:N
Stream client certificate bypass (CVE-2026-120967), 4.8 and 6.3:
  CVSS:3.1/AV:N/AC:H/PR:N/UI:N/S:U/C:L/I:L/A:N
  CVSS:4.0/AV:N/AC:H/AT:N/PR:N/UI:N/VC:L/VI:L/VA:N/SC:N/SI:N/SA:N
CVSS vector strings as printed in the F5 CVE records, read on 19 November 2026. Grouped by this site; the strings are exact.

Next the ranges for the HTTP/2 flow control flaw, which show NGINX Plus beside open source. The lines are arranged by this site from the record's version fields; "lessThan" is the record's own word, so a version listed as lessThan 1.31.9 is affected up to but not including 1.31.9.

Text
NGINX Plus     version 37.1.0   lessThan 37.3.0.1   status affected
NGINX Plus     version 37.0.0   lessThan 37.0.8.1   status affected
NGINX Plus     version R33      lessThan *          status affected
NGINX Open Source  version 1.31.0   lessThan 1.31.9   status affected
NGINX Open Source  version 1.30.0   lessThan 1.30.7   status affected
NGINX Open Source  version 1.9.5    lessThan *        status affected
module ngx_http_v2_module
Affected product versions for CVE-2026-123007 as printed in the F5 CVE record, read on 19 November 2026 (arranged by this site).

Then the workarounds. Where a record prints one, it is a configuration change that reduces exposure, not a fix. The four records that print none (the HTTP/2 flaws, the HTTP/3 flaw and the TLS 1.3 loop) leave a patch as the only remedy the saved pages print. These lines are the records' own words; check them against the F5 article before use.

Text
CVE-2026-113425  mail proxy: Restrict listeners to trusted networks.
CVE-2026-123634  resolver: Avoid using SSI or mirror directives and avoid configuring a proxied server name with variables
CVE-2026-120967  stream client certificate: To mitigate this vulnerability, configure ssl_verify_client on; in the default server of the shared listening socket, or move the stream virtual server requiring client certificates to its own dedicated listening address and port. Note: Configurations where the default server has the ssl_verify_client on; are not affected.
CVE-2026-125283  try_files: Avoid unnamed regex captures; use named captures instead. Do not reuse the same capture variable in other regex matches or for other purposes.
CVE-2026-120993  PROXY v2: Disable proxy_protocol v2; by setting proxy_protocol off; or using v1 with proxy_protocol on; OR disable abbreviated SSL handshakes by setting both ssl_session_tickets off; and ssl_session_cache off; (note: nginx built with BoringSSL or AWS-LC is not affected)
CVE-2026-123007  HTTP/2 flow control: None
CVE-2026-121810  HTTP/3: None
CVE-2026-123633  HTTP/2 use-after-free: None
CVE-2026-118899  TLS 1.3 loop: None.
Workaround text as printed in each F5 CVE record, read on 19 November 2026 (long lines wrapped by this site only where the record has line breaks).

The change log's security lines for 1.31.9, in prose: it lists ten Security items, each worded as "might". The HTTP/2 and HTTP/3 items depend on worker_connections not being enough; the resolver item needs the same name resolved at once in subrequests; the stream item says client certificates might be verified with the default server's configuration rather than the one SNI selected; the TLS 1.3 item is an endless loop while reading the request body over SSL. It also lists two behaviour changes, rejecting header lines without a colon and rejecting a request whose chunked body has too much framing overhead, and several bugfixes.

Last, the F5 article ids each record links. They are printed here as ids, not links, because this site did not read the articles.

Text
CVE-2026-123634  K000163925
CVE-2026-123007  K000163926
CVE-2026-121810  K000163927
CVE-2026-123633  K000163928
CVE-2026-118899  K000163930
CVE-2026-113425  K000163931
CVE-2026-120993  K000163932
CVE-2026-125283  K000163933
CVE-2026-120967  K000163934
CVE-2026-123640  K000163935
F5 article ids as printed in the reference field of each CVE record, read on 19 November 2026.

What to do, in order

The first item is what the pages say. The rest are this site's suggestions, ordered by exposure rather than by either rating.

Drawn from the nginx.org pages and the F5 CVE records, read on 19 November 2026. The suggestions are this site's.

Take this with you

Defender actions for the NGINX 18 November release, in order

  • What the pages say: nginx.org lists the ten as not vulnerable from 1.31.9 and 1.30.7 (the PROXY v2 entry from 1.31.9 only). Start from those two numbers.
  • Inventory every NGINX you run, including those inside container images, ingress controllers and appliances, and compare each build with 1.31.9 and 1.30.7 (this site's suggestion).
  • Check which of the ten apply to your configuration: HTTP/2, HTTP/3, mail proxy, stream, resolver with SSI or mirror, TLS 1.3, PROXY protocol version 2, alias with try_files. Patch the HTTP/2 and mail proxy ones first, because they have no printed workaround or a narrow one (this site's suggestion).
  • Ask your distribution or appliance vendor for its fixed package and the date, and do not assume a 1.30.x package carries the fix: nginx.org lists 1.30.7, and no page read says what any distribution ships (this site's suggestion).
  • Confirm ASLR is enabled on every host that runs NGINX, because the records name ASLR disabled or bypassed as the condition for code execution (this site's suggestion).
  • Do not rank by nginx.org's word alone: the five records at 8.1 and 9.2 include three that nginx.org calls medium. Rank by whether your configuration meets the record's condition (this site's suggestion).
  • Watch for worker process crashes. The record names "an NGINX worker segmentation fault"; a restart loop may be the first sign. The pages print no log line, so look for whatever your platform records when a worker exits (this site's suggestion).
  • If you use NGINX Plus, read the record or F5's article for your release line: the printed Plus ranges differ by record (this site's suggestion).

The question the two ratings leave open

If the project says medium and the vendor's record says 9.2 for the same flaw, which number decides your patch window, and could you name today the NGINX builds, the HTTP/2, HTTP/3 and mail proxy settings and the ASLR state that would settle it?

Key facts

Sources

  1. Primarynginx security advisories, read in full on 19 November 2026: the ten entries not vulnerable from 1.31.9 and 1.30.7 with their severity words and ranges, and the older entries after them. The page prints no datesnginx.org (the NGINX project; NGINX is F5's product)accessed 2026-11-19
  2. Primarynginx change log, read on 19 November 2026: Changes with nginx 1.31.9 (18 Nov 2026, ten Security items), 1.31.8 (21 Oct 2026), the other 1.31.x releases and 1.9.5 (22 Sep 2015, the HTTP/2 module)nginx.org (the NGINX project)accessed 2026-11-19
  3. PrimaryCVE-2026-123634 record, Use-after-free during response processing in resolver, read from the CVE Services API on 19 November 2026: published 2026-11-18T15:23Z, CVSS 3.1 and 4.0 metrics, affected versions, workaround and CISA-ADP SSVC entryCVE Program (assigned by F5)accessed 2026-11-19
  4. PrimaryCVE-2026-123007 record, Memory corruption in HTTP/2 flow control, read from the CVE Services API on 19 November 2026: published 2026-11-18T15:23Z, CVSS 3.1 and 4.0 metrics, affected versions, workaround and CISA-ADP SSVC entryCVE Program (assigned by F5)accessed 2026-11-19
  5. PrimaryCVE-2026-121810 record, Use-after-free in HTTP/3, read from the CVE Services API on 19 November 2026: published 2026-11-18T15:23Z, CVSS 3.1 and 4.0 metrics, affected versions, workaround and CISA-ADP SSVC entryCVE Program (assigned by F5)accessed 2026-11-19
  6. PrimaryCVE-2026-123633 record, Use-after-free in HTTP/2, read from the CVE Services API on 19 November 2026: published 2026-11-18T15:23Z, CVSS 3.1 and 4.0 metrics, affected versions, workaround and CISA-ADP SSVC entryCVE Program (assigned by F5)accessed 2026-11-19
  7. PrimaryCVE-2026-118899 record, Infinite loop while reading client request body over TLS 1.3, read from the CVE Services API on 19 November 2026: published 2026-11-18T15:23Z, CVSS 3.1 and 4.0 metrics, affected versions, workaround and CISA-ADP SSVC entryCVE Program (assigned by F5)accessed 2026-11-19
  8. PrimaryCVE-2026-113425 record, Use-after-free in mail proxy module, read from the CVE Services API on 19 November 2026: published 2026-11-18T15:23Z, CVSS 3.1 and 4.0 metrics, affected versions, workaround and CISA-ADP SSVC entryCVE Program (assigned by F5)accessed 2026-11-19
  9. PrimaryCVE-2026-120993 record, NULL pointer dereference when using PROXY protocol version 2, read from the CVE Services API on 19 November 2026: published 2026-11-18T15:23Z, CVSS 3.1 and 4.0 metrics, affected versions, workaround and CISA-ADP SSVC entryCVE Program (assigned by F5)accessed 2026-11-19
  10. PrimaryCVE-2026-125283 record, Buffer overflow when using alias with try_files, read from the CVE Services API on 19 November 2026: published 2026-11-18T15:23Z, CVSS 3.1 and 4.0 metrics, affected versions, workaround and CISA-ADP SSVC entryCVE Program (assigned by F5)accessed 2026-11-19
  11. PrimaryCVE-2026-120967 record, SNI bypass when using client certificate validation in stream, read from the CVE Services API on 19 November 2026: published 2026-11-18T15:23Z, CVSS 3.1 and 4.0 metrics, affected versions, workaround and CISA-ADP SSVC entryCVE Program (assigned by F5)accessed 2026-11-19
  12. PrimaryCVE-2026-123640 record, Insufficient validation in stream ssl_preread module, read from the CVE Services API on 19 November 2026: published 2026-11-18T15:23Z, CVSS 3.1 and 4.0 metrics, affected versions, workaround and CISA-ADP SSVC entryCVE Program (assigned by F5)accessed 2026-11-19

Share this briefing

Know someone who owns this problem? Send it to them.

Related briefings

The briefing, in your inbox

Practitioner analysis of cyber and AI security news. No vendor noise.

How often

Every new briefing in one email, at 7am, or at 7am, 12:30pm and 6pm. Nothing is sent when nothing is new. Unsubscribe any time.