Zyxel, WordPress, Apple KEV + AI Breaches: November 2026 Verified Tracker

Between November 3 and November 5, 2026, three actively exploited vulnerabilities landed in CISA's Known Exploited Vulnerabilities catalog — WordPress Core, Apple WebKit, and a Zyxel GS1900 switch zero-day — on top of a month-long wave of breaches at AI developer platforms and a widely shared OpenAI result claiming 948 working Linux root exploits in a single run. Read as a feed, it is noise. Read as a worklist, most of it is not urgent for most organizations, two items are, and at least two headlines are saying something the primary sources do not. This tracker separates the two, as of 13:17 CET on November 5, and will be updated as vendors and CISA publish more.

Developing story — first published 13:17 CET (12:17 UTC), November 5, 2026. Updated as vendors and CISA publish more.

Patch today

Ordered by exposure and confirmed exploitation. Only the first two are drop-everything items; the third is housekeeping for anyone who has not updated Apple devices since mid-2024.

1. WordPress Core — CVE-2026-119977. A time-of-check time-of-use race condition during full-package core upgrades lets an unauthenticated attacker execute code through predictable, web-accessible staging paths. It is internet-facing, exploited in the wild, and CISA set the federal due date at November 6 — tomorrow — with forensic triage required. Update to the patched Core release and leave automatic background updates on. Full analysis: our WordPress Core write-up.
2. Zyxel GS1900 — CVE-2026-107441. An out-of-bounds write in dispatcher.cgi gives a LAN-based, unauthenticated attacker OS command execution — and it survives the June patch, because that patch fixed a different bug (CVE-2026-7273). Reflash the ten affected GS1900 models to the 2.90 Patch 3 firmware; do not assume the June update covered you. If a switch was exposed, treat its configuration and root credential hashes as stolen and rotate. Full analysis: our Zyxel write-up.
3. Apple — CVE-2024-27834. A WebKit improper-authentication flaw added to KEV on November 4 (due November 18). It is a 2024 bug, CVSS 5.5, and it needs an attacker who already has arbitrary read/write to bypass pointer authentication. If your fleet is on iOS/iPadOS 17.5, macOS Sonoma 14.5, Safari 17.5, tvOS 17.5, watchOS 10.5 or later, you fixed this 18 months ago. Confirm it; do not panic over it.

Verified incident tracker

Status key: ✅ confirmed by a primary source · ⚠️ partly confirmed — the caveat is in the row · ❌ not supported by any source we could open.

Incident	What it is	Status	Fix / action	Source
WordPress Core CVE-2026-119977	TOCTOU race during core auto-upgrade; unauthenticated RCE via predictable staging paths	✅ CISA KEV · added Nov 3 · due Nov 6 · forensic triage required	Update to the patched Core release; keep auto-updates on	CISA KEV; NVD
Zyxel GS1900 CVE-2026-107441	OOB-write in dispatcher.cgi; LAN, unauthenticated OS command execution; bypasses the June CVE-2026-7273 fix	✅ Confirmed by Binarly · not on KEV	Reflash to 2.90 Patch 3 firmware (10 models); rotate switch credentials	Help Net Security
Apple CVE-2024-27834	WebKit improper authentication / PAC bypass; needs arbitrary read+write; a 2024 Pwn2Own bug	✅ CISA KEV · added Nov 4 · due Nov 18	Already fixed May 2024; ensure iOS/macOS/Safari/tvOS/watchOS on the 17.5 / 14.5 / 10.5 line or later	CISA KEV; Cyber Security News
AI developer-platform breaches ("Black October")	≥9 platforms breached over October; attackers hunt stored environment variables holding AI keys → LLMjacking	✅ Confirmed by Wiz	Rotate AI inference keys at the provider (see below)	Wiz
WorkOS breach (Oct 30)	Internal sales application accessed via an AI-assisted offensive framework; CRM and sales-call data exposed	⚠️ Confirmed breach, but no secrets exposed per WorkOS; not part of Wiz's LLMjacking list	Watch for targeted phishing; vendor requires no key rotation	WorkOS bulletin
OpenAI "948 Linux root exploits"	Research pipeline turned kernel bugs into 948 distinct local privilege-escalation exploits in one run	✅ Confirmed as research, not an attack	Fast-lane kernel LPE patching; shrink the local-account surface	Help Net Security
CERT-In "Crafty Falcon" / ARTEX (India BFSI)	Reported AI-driven actor targeting Indian banking; backdoored ARTEX Docker image, ~62 consoles exposed to root	⚠️ Single outlet; no primary CERT-In advisory we could open confirms the attribution or the figures	Hunt for internet-exposed ARTEX consoles; treat open offensive-AI tooling as an exposure	ARTEX tooling corroborated — see related post

Five of the seven are confirmed against a primary source. Two carry caveats, and the caveats are the point: a tracker that lists everything at the same confidence is as useless as the viral thread it is trying to correct.

Rotate these secrets

The AI developer-platform wave is a credential-theft story, not a code-theft story. Wiz, which named the pattern "Black October" for AI developer platforms, counts at least nine platforms — Fireworks AI, Andromeda, Zeabur, Lightning AI, OpenCode, Railway, Warp, CommandCode, Northflank — with Trigger.dev, Brev and Chainstack added in the first days of November. Across the reports the common thread is explicit: attackers hunt stored environment variables, with a focus on those that grant AI access. Wiz traced two enterprise cases where stolen keys ran up over $100,000 in inference usage before anyone noticed.

If you held any credentials on a platform named above, the rotation checklist is short and the order matters:

• Rotate at the provider, not in the dashboard. Deleting the exposed secret inside the breached platform does not revoke it. The key keeps working until you roll it at OpenAI, Anthropic, AWS Bedrock, Google, or wherever it actually authenticates.
• Rotate every key that platform could read, not just the one you remember pasting. Build pipelines and agent consoles accumulate forgotten tokens in environment variables.
• Check provider usage dashboards for the breach window for spend you did not generate, model calls from unfamiliar regions, or sudden quota exhaustion — the signature of resale-driven LLMjacking.
• Scope the blast radius of each key: a single over-privileged inference key is often also a billing and fine-tuning key.

WorkOS is the counter-example, and it is why this section is not simply "every AI breach means rotate your keys." WorkOS disclosed an October 30 intrusion in which an AI-assisted offensive framework automated the reconnaissance and exploitation, but the data exposed was CRM and sales-call material — not API keys, and not production WorkOS data. The right response there is phishing vigilance for named contacts, not a key-rotation fire drill.

What the viral lists got wrong

Three claims from this week's roundups do not survive contact with the primary sources.

"Apple zero-day under active attack — patch now."

CVE-2024-27834 is not a 2026 zero-day. It is the WebKit bug Manfred Paul demonstrated at Pwn2Own Vancouver in 2024, fixed by Apple in May 2024. CISA added it to KEV on November 4, 2026 because exploitation is now observed in the wild — but it is CVSS 5.5, local, and it presumes an attacker who already holds an arbitrary read/write primitive. For a fleet that has taken any Apple update since mid-2024, this is already closed. It belongs on a verification checklist, not at the top of an incident bridge.

"An AI just wrote 948 Linux root exploits — attackers now have 948 new weapons."

The real result is narrower and more interesting. OpenAI's pipeline produced 1,169 working root exploits, which grouped into 948 distinct vulnerabilities — the headline number is a deduplicated bug count, not a weapon count. They are local privilege-escalation exploits: an attacker needs a foothold on the box first. And the run cost $1,213,430 in API credits. The honest lesson is not "948 weapons are loose"; it is that reproducing and triaging known-bug-class findings at scale is now cheap enough to matter, and that even OpenAI's own early run over-counted — 20 of 200 claimed successes had reached root through a different bug than the one assigned. Our full breakdown covers the economics.

"WorkOS, Northflank and Trigger.dev — same LLMjacking wave."

Northflank and Trigger.dev belong to Wiz's AI-developer-platform wave, where stored AI keys were the prize. WorkOS does not appear on Wiz's list, and its own bulletin describes a CRM/sales-data breach with no secrets exposed. Same month, same attacker tradecraft (AI-assisted frameworks), different incident and different remediation. Grouping them pushes WorkOS customers toward a key rotation they do not need and away from the phishing watch they do.

How to prioritize when everything is critical

A week that stacks three KEV entries on top of a breach wave is exactly where a flat "patch all criticals" policy burns a team out on the wrong work. CISA itself tiered this batch: WordPress got a three-day window, Apple got fourteen. That is BOD 26-04 risk-based patching doing its job — of the 137 KEV additions between June and October 2026, 111 were assigned the shortest tier, so "it is on KEV" no longer means "drop everything."

The discriminator is not severity, it is reachability times exploitability:

• Internet-facing and exploited → today. WordPress Core and exposed Zyxel switches qualify.
• Exploited but precondition-heavy or local → this cycle, not this hour. Apple CVE-2024-27834 and the OpenAI-class kernel LPEs need an attacker already inside.
• Validate before you patch in a panic. A banner version alone did not tell Zyxel owners whether they were exploitable — the June "fix" was present and the hole was open. Confirm the actual reachable condition before you schedule an emergency change window.

Track the KEV deadlines against your own estate on the CISA KEV tracker, and the vendor-specific picture on the Zyxel and Linux kernel hubs.

Where this leaves you

The unifying thread this week is not any single CVE — it is that the attacker's workflow is increasingly machine-paced. Offensive AI frameworks automated the recon in the platform breaches; a research pipeline turned a kernel into 948 exploit candidates for the price of a mid-size pentest. The defensive answer is not to read faster. It is to run the same machine-speed offense against your own estate first, continuously, and to let the output be a ranked fix list rather than another feed to triage.

That is what Zero Hunt is built to do. Its AI Remediation Advisor ranks findings by real exploitability — CISA KEV first, then what proved exploitable in your own campaign, then CVSS and EPSS — so a week like this one resolves to "these two, in this order," with the exact fixed version, a rollout-and-rollback note, and a re-verification using the same proof that demonstrated the issue. Underneath it is an autonomous AI red team that runs on an on-premise appliance on its own private models, with a human approving every action that matters, so the offense you point at your WordPress fleet, your edge switches and your Linux hosts never leaves your network and never bills per token. The attackers this week were not smarter than your team. They were faster, and they only had to be right about which door was open.