This is the fourth issue of the NT IT security weekly, covering week 46 of 2026 (from 9 November). Data was read up to the early hours of 16 November. This issue was started on 13 November, and before publication we re-read the fields that change — two of them really had changed, and both moved in the helpful direction. Throughout, we mark which sentence comes from the 13 November reading and which from the 16 November re-read.
The one thing worth doing today, first: if your site runs Apache HTTP Server, you can now fix it. When we reported this critical last week, Debian's stable releases had nothing to upgrade to. On 12 November both Debian and Amazon Linux shipped 2.4.71, so the "we know, but there is nothing we can do" state ended this week.
The second story is this week's highest score: CVE-2026-126187 in React Server Components, 9.5 under CVSS 4.0, unauthenticated remote code execution. But the official description itself adds three qualifiers, and quoting the 9.5 without those three sentences uses a true number to create a false impression. The three sentences are quoted in full in the first section.
The third story is the only entry this week that the US CISA treats as actively exploited: OpenSSH, CVE-2026-123158. It entered the catalogue on 9 November, and the deadline CISA set for federal agencies was 12 November — already past. The conditions for this one are narrow; how narrow is spelled out in section three.
How to read this weekly
- Every item is tagged by source layer: [Official] is the vendor, project or government agency itself; [Research/Media] is a security firm or a news outlet, and always comes after the official line. Where the two disagree, we print both and do not pick a winner.
- Status fields always carry a date. Identifiers, affected ranges and fixed version numbers do not change, so we state them plainly. Sentences such as "no score yet" or "no fixed version yet" have a shelf life, so we write "as of <date>". Two such sentences expired within three days in this issue, and we say below exactly which two.
- Fixed version numbers are copied from the official advisory, never from a second-hand roundup. Where a count is ours, we say it is ours.
- You have to know which versions you are running before any of these numbers mean anything. How to check, and where it is easiest to get it wrong, is in Security Basics 01|An advisory just landed: how do you know whether it affects you.
- Where advisories live, and who should be reading them, is in Security Basics 06: Where to read security advisories, and how to confirm someone is watching them.
1. React's 9.5: the number is real, and the project wrote three qualifiers around it
[Official: GitHub, the CNA for this record] CVE-2026-126187 affects three packages: react-server-dom-webpack, react-server-dom-parcel and react-server-dom-turbopack. The official description reads that these packages "are vulnerable to unauthenticated remote code execution through crafted payloads targeting frameworks that support React Server Components." The fixed versions are 19.0.9, 19.1.10, 19.2.9 and 19.3.1, with the affected range starting at 19.0.0.
Where the score comes from matters, because it decides whether the number can carry a decision. Reading NVD in the early hours of 16 November: the status is Deferred, last modified 13 November, CVSS 4.0 base score 9.5 (CRITICAL), vector CVSS:4.0/AV:N/AC:L/AT:P/PR:N/UI:N/VC:H/VI:H/VA:H/SC:H/SI:H/SA:N, and the scoring source is [email protected] — GitHub, which is the CNA for this record. So this is not "NVD scored it 9.5"; it is "the CNA scored it 9.5 and NVD carries the record." Neither is more credible than the other, but they are different sources, and saying it the wrong way attributes somebody else's judgement to NVD.
Three qualifiers, written by the project in the same paragraph
The same official description continues with three sentences, each of which narrows the scope:
- "The vulnerability only affects extremely rare cases where responses transport multiple typed arrays and an attacker-controlled value."
- "Applications whose React code does not use a server are not affected."
- "Applications that do not use a framework, bundler, or bundler plugin that supports React Server Components are not affected."
In plainer terms: if your React runs only in the browser — a pure front-end single-page app, a statically exported site — the project says you are not affected; and if you do not use a framework or bundler that supports React Server Components, the project says the same. What is exposed is server-side React that actually uses RSC, typically through a framework such as Next.js.
So why write it up at all? Because "do we use RSC" is usually not something the owner of the business knows — it is something whoever built the system knows. The check is two steps: ask whether the front end is a framework such as Next.js, then have someone look at package.json to see whether React is at 19.0.0 or above and whether it has moved to one of those four fixed versions. If neither question can be answered, that is itself the finding — it means nobody can currently describe what this system is made of.
One more thing about this item, stated openly: it is not on our standing source list. We first saw it in a second-layer source — a security firm's blog — and that firm wrote that it had not checked with React itself. So every number and every quotation above was re-read by us against NVD and the official advisory (GHSA-xqc2-xf4c-x3h3). Second-layer material does not go into this weekly unchecked.
2. Apache HTTP Server: this week it went from "no score and nothing to upgrade to" to "scored, and the upgrade path is open"
[Official: Apache] The advisory title for CVE-2026-119618 reads mod_http2: heap corruption and RCE in assign_headers, and the affected range reads This issue affects Apache HTTP Server: from 2.4.55 through 2.4.70. The fixed version is 2.4.71, released 11 November. The "critical" rating is Apache's own.
The score: absent on 13 November, present on 15 November
We read NVD on 12 and 13 November, and on both days the metrics block was empty — there was no CVSS score at all. Reading again in the early hours of 16 November, the status is Modified, last modified 15 November, and there is now a CVSS 3.1 score of 8.1 (HIGH), vector CVSS:3.1/AV:N/AC:H/PR:N/UI:N/S:U/C:H/I:H/A:H.
Who produced that 8.1 belongs in the sentence. NVD marks it Secondary and identifies the source by a UUID; we looked that UUID up in NVD's own source list, and the organisation name is CISA-ADP (the US CISA acting as an Authorized Data Publisher). So the accurate statement is: Apache rates it critical, CISA-ADP has added an 8.1 HIGH, and NVD itself has not finished its analysis. Those are three facts; compressing them into "NVD scored it 8.1" drops two of them.
Note also that 8.1 looking lower than "critical" is not a contradiction of Apache: the HIGH band under CVSS 3.1 tops out at 8.9, and the AC:H term — high attack complexity — pulls the number down. When you are scheduling work, read the vector, not the name of the band.
The upgrade path opened on 12 November
[Official: Debian] On 13 November, reading Debian's security tracker row by row, the security suites for both bookworm and trixie still carried the old version and were marked vulnerable. Re-reading in the early hours of 16 November, both lines are fixed:
| Suite | Version | State |
| bookworm (base) | 2.4.68-1~deb12u1 | vulnerable |
| bookworm (security) | 2.4.71-1~deb12u1 | fixed |
| trixie (base) | 2.4.68-1~deb13u1 | vulnerable |
| trixie (security) | 2.4.71-1~deb13u1 | fixed |
| forky, sid | 2.4.71-1 | fixed |
The tracker names DSA-6620-1 (trixie) and DLA-4902-1 (bookworm). We opened both announcement mails ourselves to confirm the dates: both are 12 November 2026.
The two base rows still showing the old version is normal, not "unfixed" — Debian ships security fixes through the security suite rather than rewriting the base suite's version. So "have we fixed it" cannot be answered by checking whether apache2 -v prints 2.4.71; it depends on whether the machine has the security suite enabled and has actually run an update.
[Official: Amazon Linux] Amazon was slightly earlier: on 12 November it issued two critical advisories, both for httpd, both covering this CVE — ALAS2023-2026-3458 (fixed package httpd-2.4.71-1.amzn2023.0.1, with the official instruction reading Run dnf update httpd --releasever 2023.12.20261112) and ALAS2-2026-4114 (fixed package httpd-2.4.71-1.amzn2.0.1, official instruction Run yum update httpd, noted as applying to the AL2 Core repository). Those were the only two advisories Amazon Linux rated critical this week.
It is not in the exploited catalogue
[Official: US CISA] In the early hours of 16 November we pulled the full KEV data file (version 2026.11.13, 1,781 entries) and checked entry by entry: CVE-2026-119618 is not listed. "Scored high" and "being exploited" are two different things, and this one currently has only the first.
3. OpenSSH: the only entry this week in the exploited catalogue, and its deadline has passed
[Official: US CISA] CVE-2026-123158, vendor OpenBSD, product OpenSSH, added 9 November with a 12 November deadline, knownRansomwareCampaignUse set to Unknown. KEV lists what there is evidence of being exploited in the wild — we checked every other item in this issue (React, Apache, Ivanti) against the same file, and this is the only one that is listed.
[Official: NVD/MITRE] Status Analyzed, published 5 November, CVSS 3.1 of 8.1 (HIGH), scoring source [email protected]. The official description reads: "sshd in OpenSSH before 10.6p2 on Linux, when musl libc is used without PAM, lacks a workaround for a password authentication bypass … For some configurations of accounts, any password between 256 and 1024 characters succeeds. An example of a platform using musl libc without PAM is Alpine Linux." The fixed version is OpenSSH 10.6p2.
The conditions are the story, so here they are in full: musl libc, and no PAM, and sshd actually running, and an account configuration in the state the description names. The example the project gives is Alpine Linux.
One sentence we deliberately do not write: "distributions using glibc — Debian, Ubuntu, Amazon Linux — are safe." The description states the conditions under which the flaw holds; it does not state the converse. Inverting the conditions into a clean bill of health is the most common way to misread this kind of advisory, and it fails in the direction that makes people relax.
For most companies the real hook is containers. Alpine is the base of a great many Docker images, so the first reaction is often "we use Alpine, we are in trouble". But the question is not whether you use Alpine; it is whether anything inside that container is running sshd. Most Alpine images do not ship sshd, so the real exposure is much narrower than the headline. The typical hit is the container where somebody installed ssh "so we can get in and look around" — which is worth reviewing on its own merits.
There is one claim we could not trace to a primary source, so we leave it out. Media coverage quotes a security firm saying exploitation began within hours of the fix being published. We could not find that firm's own text, so this weekly does not repeat it. We have a harder substitute: CISA listed it on 9 November with a 12 November deadline, and that is the agency's own statement.
4. Node.js, 11 November: three release lines at once, highest severity HIGH
[Official: Node.js] The security release of 11 November reads: "Updates are now available for the 26.x, 24.x, 22.x Node.js release lines" and "The highest severity issue fixed in this release is HIGH." The fixed versions are v22.23.5, v24.22.3 and v26.13.1.
Counting the entries on the official page ourselves: 7 high, 6 medium, 3 low — 16 in total. That count is ours; the page does not publish a total.
The ones worth singling out (titles are the project's own wording):
- CVE-2026-123863 (high):
HTTP/2: remote abort, CHECK_EQ(current_nghttp2_memory_, 0) in ~Http2Session() — the description reads "A reachable assertion in Node.js HTTP/2 session teardown lets any unauthenticated client abort the server, over both TLS and cleartext." No account, no credential, and the service goes down.
- CVE-2026-123864 (high):
HTTPS: a zero-length TLS write re-enters llhttp_execute(), and the server copies its own heap into the request body.
- CVE-2026-123872 (medium):
TLS 1.2 renegotiation allows clients to change servername unilaterally to bypass client certificate authentication.
- CVE-2026-123871 (medium):
Proxy credentials disclosed in ERR_PROXY_INVALID_CONFIG error messages.
The release also carries dependency updates: llhttp (22.x/24.x/26.x), undici (24.x/26.x) and sqlite (22.x/24.x). That matters for scheduling: even if none of the identifiers above applies to you, the dependency fixes ship inside the same version, so there is nothing to separate out.
One thing we did not confirm, stated as such: the official page also shows CVE-2026-48615 and CVE-2026-48937, which are not among the 16 titles we counted. They may correspond to those dependency updates, but we did not verify that, so we do not state it as a conclusion.
A related example of how dependency security updates travel quietly: Electron released v45.3.3 on 14 November. Its release notes mention one DevTools fix and the line Updated Node.js to v24.22.3. — and v24.22.3 is exactly the 24.x fixed version from the Node.js security release above. Electron did not label this release a security update; the two lining up is something we saw by putting the two official texts side by side, not something either side announced. If you filter advisories by whether the title says "security", this one slips past.
5. Composer: four security fixes, two of them 8.8
[Official: Composer] On 10 November, 2.11.2 and 2.2.32 (LTS) shipped together. The severities are the project's own:
- GHSA-jhfp-4ppm-crmm, high 8.8 — the release notes read "Fixed arbitrary command execution via the
prevent_url_access_callable transport option being loaded from composer.lock / vendor/composer/installed.json". In other words, a tampered composer.lock is enough to make the install step run arbitrary commands.
- GHSA-w6hj-6587-3jh5, high 8.8 — "Fixed malicious zip archives being able to write files outside of the package directory via symlinked entries when extraction falls back from
unzip to ZipArchive". Note "falls back": it applies only when unzip is absent and Composer uses PHP's own ZipArchive — and minimal container images are exactly where unzip tends to be missing.
- GHSA-x82q-x274-r6pq, medium 6.5 — "Fixed inline repository HTTP credentials being sent to dist mirrors on other hosts supplied by a crafted
composer.lock". Private-repository credentials sent to somebody else's host.
- 2.11.2 additionally updates the bundled
symfony/filesystem to address GHSA-6prc-h956-hxjq. That advisory was published by Symfony on 4 November and belongs to the previous issue, not this week.
The two releases do not fix the same set, and that is easy to miss: 2.2.32 does not include GHSA-jhfp-4ppm-crmm, nor the symfony/filesystem update. If you are on the 2.2 line, upgrading is not the same as being fully covered.
All three advisories have an empty CVE field. Whether that is "not yet assigned" or "will not have one" is not stated — so any tooling that filters on the presence of a CVE identifier misses this whole batch.
6. Laravel: two items, low scores, but they hit "temporary links"
[Official: Laravel] Two advisories on 11 November. Both are medium, both CVSS 3.1 of 4.8, and both have an empty CVE field. The fixed versions are the same for both: 12.72.0 on the 12.x line, 13.41.0 on the 13.x line.
- GHSA-v5ch-g375-82r6: "When PHP's
max_input_vars limit causes the expires query parameter to be dropped from the parsed query, Laravel's signed URL validation can still verify the signature against the raw query string. signatureHasNotExpired() then treats the URL as non-expiring, so an expired temporary signed URL may be accepted." In plain terms: a link that was supposed to expire may work forever. The project also gives a mitigation: "Limit URL length at the web server. Default nginx and Apache settings already reject the long URL the attack needs." — so the sites actually at risk are those that raised the URL length limit.
- GHSA-6rv8-qj5v-p8ch: "When an application passes attacker-influenced, numeric-keyed parameters … to
temporarySignedRoute() or signedRoute(), and those parameters do not fill a route placeholder, Laravel appended their values to the query string without URL encoding."
Why a low score still earns a section: "temporary links" are what a lot of systems use for password resets, file downloads and external sharing. The risk in a "temporary" link that never expires is not in the score; it is in what the link points at.
7. LiteLLM: narrow conditions, but hitting them means the whole proxy
[Official: LiteLLM] GHSA-2rwg-2wh4-7c4f, published 12 November, high, CVSS 3.1 of 7.7, CVE field again empty. Fixed in 1.106.9, 1.108.6, 1.109.1 and 1.110.0-rc.2.
The project states the conditions narrowly: "This vulnerability affects only deployments with at least one Azure OpenAI (azure/) or Azure AI (azure_ai/) model, and which do not have AZURE_API_KEY or AZURE_OPENAI_API_KEY environment variables set."
What follows when they hold is also stated: a low-privilege authenticated user can read the proxy process's environment variables and the contents of files those variables point at — "On a typical deployment this includes the database connection URL, provider API keys and LITELLM_MASTER_KEY. Reading the master key gives full proxy admin control."
It is in this issue for its shape, not its score: more and more companies route every LLM call through a single proxy, and that proxy usually holds all the keys. Centralising is the right call, but it also turns "one person got in" into "everything is readable".
8. Ivanti Endpoint Manager: two items, and the scores are the vendor's own
[Official: Ivanti] Two advisories on 10 November, both affecting 2024 SU7 and earlier, fixed in 2024 SU8:
- CVE-2026-121653: SQL injection in Core, unauthenticated remote code execution, CVSS 3.1 of 9.8 (CRITICAL).
- CVE-2026-18128: the Agent transmits sensitive information in cleartext; an attacker has to be in a man-in-the-middle position. 6.8 (MEDIUM).
The source of those scores belongs in the sentence. Reading NVD in the early hours of 16 November: both records are Awaiting Analysis (on 13 November both still read Received, so they moved one step in three days), and the scoring organisation, per NVD's own source list, is ivanti — the vendor itself. So this is not "NVD scored it 9.8". Ivanti's own advisory writes the vector as CVSS:3.0 while NVD carries 3.1; the number is the same and only the version differs, but we note it.
Ivanti also states it is "not aware of any customers being exploited", and we confirmed on 16 November that neither record is in KEV.
Why most of our readers can skip this, and why we print it anyway: Ivanti EPM is enterprise endpoint-management software that smaller companies generally do not run. But KEV already holds 6 Ivanti EPM entries, among them CVE-2024-29824 — also a SQL injection in Core, added 2 October 2024 — while CVE-2021-44529 carries Known in the ransomware column. Same product line, same shape of flaw, already used in the wild: that is a reason to take this one seriously, and it is a record rather than our opinion.
9. CISA's exploited catalogue: 12 entries this week, two of them due today
[Official: US CISA] The data file we pulled in the early hours of 16 November is version 2026.11.13 (released 13 November, 20:24 UTC) with 1,781 entries. Twelve were added between 9 and 15 November, and all twelve carry Unknown in the known-ransomware column.
| Added | Identifier | Product | Due |
| 11-09 | CVE-2026-102895 | GPUStack | 11-12 |
| 11-09 | CVE-2026-116779 | SignalWire FreeSwitch | 11-12 |
| 11-09 | CVE-2026-123158 | OpenBSD OpenSSH | 11-12 |
| 11-09 | CVE-2026-123218 | IBM Aspera Faspex | 11-12 |
| 11-10 | CVE-2026-111560 | Barracuda (multiple products) | 11-13 |
| 11-10 | CVE-2026-116606 | Barracuda Email Security Gateway | 11-13 |
| 11-10 | CVE-2026-111793 | Openwave Webtop Media Transcoder | 11-24 |
| 11-10 | CVE-2026-124194 | Microsoft Office | 11-24 |
| 11-10 | CVE-2026-125350 | Google Chromium | 11-24 |
| 11-10 | CVE-2026-20320 | Cisco BroadWorks | 11-24 |
| 11-13 | CVE-2026-114713 | UnrealIRCd | 11-16 |
| 11-13 | CVE-2026-123039 | Check Point (multiple products) | 11-16 |
The two added on 13 November are due today, 16 November. KEV deadlines bind US federal agencies and are not a legal obligation for companies in Taiwan; but they are a judgement worth borrowing — when the agency gives its own people three days, it is saying this cannot wait until next month.
One window we cannot see, stated plainly: at the time of reading, the data file was still at version 2026.11.13. So whether anything was added on 14 or 15 November is invisible to this issue — not "nothing was added", but "the file has not been updated". This is the same shape as issue 44, where the file only went up to 30 October.
[Official: CSA Singapore] On the same day, 13 November, Singapore's Cyber Security Agency issued AD-2026-017, "Advisory on Recent Uptick in Reported Cyber Incidents", which contains the line "The growing use of Artificial Intelligence by threat actors could further accelerate the discovery and exploitation of vulnerabilities". The same advisory also carries a sentence about reported incident counts, but we did not read that sentence through to the end, so we do not quote it here. It is listed here because it happened this week, not because we verified its statistics.
10. Browsers and mail clients: Mozilla's 10 November batch
[Official: Mozilla] Five advisories on 10 November: Firefox 160 (MFSA 2026-125), Firefox ESR 153.7 (2026-127), Firefox ESR 115.45 (2026-126), Thunderbird 160 (2026-128) and Thunderbird 153.7 (2026-129); plus Firefox for iOS 160.0 (2026-124) on 9 November.
The Firefox 160 advisory carries an overall impact of high, and its first entry reads "CVE-2026-124852: Incorrect boundary conditions in the Graphics: CanvasWebGL component". We counted 99 distinct CVE identifiers on that page, but that is a string count, not a verified entry count, so please do not read it as "99 holes fixed".
The two ESR lines deserve a separate note: many companies pin their standard desktop build to ESR, and ESR version numbers do not line up with the mainline. So "our Firefox is 153.7, which is older than 160, we must be behind" is the wrong conclusion — 153.7 is the fixed version on the ESR line.
11. Checked and found nothing, and what we did not check
This section exists because "we did not see anything" and "we did not look" are different, and a reader who cannot tell them apart will assume the first.
Checked, no security advisory this week (9–15 November):
- PHP — the official archive shows only 8.6.0RC7 on 12 November inside the window, and a release candidate is not a security release.
- WordPress — the security category of the official news stops at 7.1.4 on 28 October; the four posts inside the window are not security releases (the 7.2 Beta 5 post of 10 November states
testing and development only). Core advisories inside the window: zero.
- Symfony — both the advisory category and the repository advisories stop at 4 November, which belongs to the previous issue.
- Next.js — advisories stop at the 3 November batch.
- Sansec — the research source we added from this issue onward; the most recent post is 26 October.
- AWS security bulletins — nine published (2026-174-AWS through 2026-182-AWS, 9 to 12 November), but all nine concern AWS's own SDKs, CLIs and developer tooling (ParallelCluster CLI, Braket SDK, API MCP Server, Deadline Cloud worker agent, Kiro CLI, mcp-gateway-registry twice, SDK for pandas, Advanced Go Wrapper). None concerns a cloud service itself, and none of the summaries carries a severity.
What we did not check, also stated:
- WordPress plugins and themes — a different order of magnitude of source, and not on our standing list. In practice, when a WordPress site is breached, plugins account for far more of it than core does. This is a gap in this weekly, not a "nothing this week".
- Microsoft's November rollup — we did not read Microsoft's own update guide. The "how many were fixed at once" figure circulating this week was counted by a security firm; we did not check it against Microsoft's own page, so we do not repeat it here.
- EPSS (the exploit-probability score) — not checked at all this issue.
The one sentence to take away
Scores are for ordering work, not for frightening people. This issue contains a 9.5, a 9.8 and two 8.1s — yet the one that needs attention today (OpenSSH) is not the highest-scored, and the highest-scored one (React) does not apply to most readers at all. What separates them is not the number: it is those three qualifiers, that vector string, and whether anyone has actually used it.
This issue also demonstrates something else: the same vulnerability looks different three days apart. On 13 November Apache was "no score, and nothing to upgrade to". On 16 November it is "scored, and two major distributions have shipped". If nobody is watching these advisories for your systems on a regular basis, what you see is always one day's snapshot — and that day is usually the day something went wrong.
If, having read this issue, you cannot answer "which versions are we running, who reads the advisories, and when did we last update" — that is not a security problem. It is a system maintenance problem that has not been assigned to anyone yet. Our system health check answers those three questions first: establish what is actually there, then talk about long-term maintenance.