The Briefing

What happened, why it matters, and what to do about it. Analysis, not rewritten press releases.

Threat Intel9 Sept 2026

Two Tanium report permissions can become code execution. Three release branches need checking.

Comply has two distinct access paths in these advisories. The fixed module versions matter more than a platform-level update checkbox.

An empty out-of-hours IT operations desk during a night maintenance window: two dark switched-off monitors, a keyboard pushed askew, a mug of tea with a spoon, a clipboard holding a printed checklist with lines ticked in biro, headphones and a notepad, lit by a single desk lamp, with a server rack of amber and green standby lights visible through a glass partition behind.

Threat Intel8 Sept 2026

Two perfect scores in this Patch Tuesday need no action. The two being exploited score 7.8.

Read from Microsoft's own CVRF. Nothing is publicly disclosed, the 1,170 figure is a month's accumulation, and the field to check is Customer Action Required.

A small online retailer's despatch bench at night after closing: three sealed plain cardboard parcels stacked under a caged work lamp, a small thermal label printer with a curl of blank labels spilling onto the bench, a black card payment terminal with a dark screen, a roll of packing tape and scissors, with racking of plain stock cartons receding into shadow behind.

Threat Intel8 Sept 2026

Adobe told Commerce merchants to rotate their payment gateway keys. The coverage stopped at the patch.

Read from Adobe's own bulletin, the NVD record and CISA's catalogue. A fully patched store was still taken, and the catalogue has not moved since Friday.

A small unbranded network router with external aerials on a steel bench at night, its lid removed to expose a green circuit board, yellow and blue patch leads looping from its ports, an amber status light lit, with a blurred terminal display behind it.

Threat Intel7 Sept 2026

A national CERT found six RouterOS flaws with a cyber model, and wrote down exactly what the model did not do

MikroTrick, the indicators of compromise, and the methodology section almost nobody quoted. The vendor's own detection marker does not prove a device is clean, and CERT Polska say so.

An unbranded rack-mounted network appliance pulled half out of a dark server rack on its slide rails, network cables hanging loose beneath it, lit by a hard torch beam from the right.

Threat Intel7 Sept 2026

SonicWall's exploited SMA 1000 flaws need four remediation steps, and almost everyone reported one

Vendor-confirmed exploitation, a KEV deadline that expired on Saturday, and a remediation procedure that assumes you may already be compromised.

A grid of dark stained wooden library card catalogue drawers photographed at a raking angle, each with a tarnished brass label frame catching the light. Every label frame is empty.

Threat Intel5 Sept 2026

Cisco found these with frontier AI models, grouped them into seven umbrella CVEs, and never said how many bugs there are

The most consequential sentence in the IOS XR advisory is the one about how the flaws were found, and almost every write-up removed it. What follows from it is a counting problem.

A pair of heavy brushed-steel doors standing almost closed in a dark concrete room, with a narrow vertical sliver of hard white light spilling through the gap and laying a thin bright wedge across the floor.

Threat Intel5 Sept 2026

The KEV remediation window fell from 21 days to three, and the 60-day tier everyone quotes has never existed

A directive reported as reducing the patching burden coincides with the most aggressive tier becoming the default on three quarters of new entries. Two opposite errors are circulating about it.

A single deep crimson wax seal, freshly pressed and glossy, on a sheet of near-black paper, lit by a cold white rim light from the right. The seal is smooth and unstamped, carrying no crest, monogram or mark of any kind.

Threat Intel3 Sept 2026

A BGP hijack delivered a malicious Virtualizor update, and the TLS certificate on it was genuine

AS62390 announced a Hetzner /24 for 33 hours, diverting Softaculous update traffic. Let's Encrypt issued a valid certificate because its own domain validation was routed through the hijack.

A single small blank white index card lying on a vast matte charcoal surface, lit by a cold rim light from the right, the left half of the frame falling away into near-black. The card is entirely unprinted.

Threat Intel3 Sept 2026

The whole public evidence base for this morning's Falcon zero-day is 744 characters, and two of its sentences did not survive the retelling.

Exploit code for an unpatched local privilege escalation in CrowdStrike Falcon Sensor appeared at 02:48 UTC. The README contains two qualifications the coverage has dropped.

A machined dark aluminium cube sectioned open to reveal a second shell inside it, then a third, and at the centre a small crimson-lit core, on a dark surface.

Threat Intel2 Sept 2026

CISA catalogued three exploited flaws this evening that nobody has reported. Two of them cannot be found by vendor name.

Kestra, LiteLLM and Starlette joined the KEV catalogue at 16:54 UTC with no coverage at all. All three were months old, and the titles misdescribe two of them.

An ornate brass keyhole escutcheon on a dark aged wooden door, with a thin beam of warm light escaping through the open keyhole into the darkness, lit by a crimson rim light.

Threat Intel2 Sept 2026

A Chinese-speaking crew turned Brazilian .gov sites into SEO infrastructure. Their Apache module strips the CSP on three paths and leaves the rest correct.

Check Point Research documents Gambling Goblin compromising Brazilian government web servers since mid-2025. The module reverse-proxies three hardcoded prefixes and strips their Content-Security-Policy.

An antique brass klaxon alarm horn lying on a dark surface, its flared bell facing the camera, lit by a crimson rim light with the left half of the frame falling away into near black.

Threat Intel2 Sept 2026

SonicWall, JFrog and Langflow were reported exploited on one source each. CISA catalogued two of them eight hours later.

Read from the KEV feed and the NVD API on 2 September 2026: three loud exploitation claims, none catalogued, and a substitution trap where searching KEV by product name clears the wrong identifier.

A small sealed brown cardboard shipping carton on a dark matte surface, its shipping label torn away leaving a pale rectangle of adhesive residue, lit by a cold indigo rim light.

Threat Intel1 Sept 2026

Packagist removed thirteen malicious Composer themes. A fourteenth from the same operator is still listed.

Thirteen trojanised theme packages were pulled after Socket published. Checked by hand today: a sibling package Socket named as a sleeper is still installable, and the malicious jQuery is still on GitHub.

A blank manila envelope on a dark wooden desk, one corner slightly lifted to reveal darkness inside, lit by a cold indigo rim light.

Threat Intel1 Sept 2026

A fake coding test forbade AI assistants. The researchers say an AI would have caught the malware.

Kaspersky documents a recruitment lure whose README bans AI code review, declares the trojanised file bug-free, and ships a malicious package inside the archive so no scanner ever sees it.

A brass key blank lying on a dark workbench beside a pile of identical uncut blanks, lit by a cold indigo rim light against black.

Threat Intel1 Sept 2026

A 9.8 JFrog Artifactory bypass was exploited four days after disclosure. The patch shipped on day one.

CVE-2026-82329 lets an unauthenticated attacker reach admin on Artifactory. JFrog fixed it the same day it was published, its advisory misstates the 7.146 fix, and the exploitation claim rests on one firm.

A single strip of yellow and black striped hazard tape stretched across a dark open doorway, with the empty unlit room clearly visible beyond it.

Threat Intel1 Sept 2026

UAC-0099 hid a nuclear weapons essay in malware to stop AI analysis. Nobody has tested whether it works.

ESET disclosed an inert comment block designed to make AI analysis refuse. The whole primary source is three posts on X, nobody tested it, and Endor Labs documented the same trick in June.

Two identical blank sheets of heavy cream paper on a dark desk, the top one curling to reveal the copy beneath, captioned DoJ corrected one copy, the other still says victims, one of two, and 17 articles never corrected at all

Threat Intel31 Aug 2026

DoJ corrected its China hacking claim. It fixed one copy of two, and 17 articles still carry the original

A correction that reaches one of two identical documents has not been made. Both copies are live on justice.gov and they say different things.

A single glass tick box glowing cyan on a black background, captioned: eight stages in, not one is a CVE. 0 CVEs, no patch exists, because nothing is broken.

Threat Intel30 Aug 2026

TerminalFix chains eight stages from a fake CAPTCHA to a reverse tunnel, and never touches a CVE. There is nothing to patch

Every stage uses documented Windows behaviour, so there is no patch, no KEV deadline and no scanner finding. The controls that reach it had to be set before the lure arrived.

A brass pipe T-junction with one branch sealed by a blank cap, macro close-up against dark industrial pipework, the pipes edge-lit in cyan and the capped branch catching crimson light.

Threat Intel17 Aug 2026

Four CVEs let agent tools run without the model running at all. The guardrails were not bypassed, they were never asked

Black Hat 2026 research with four identifiers, one unpatched open-source path AWS has assigned to the customer, and the reason prompt hardening measured the wrong component.

A recessed circular power button ringed in crimson light on the dark brushed-metal front panel of a rack-mounted server, macro close-up, with a cyan edge light down the left side of the chassis.

Threat Intel14 Aug 2026

An Akira affiliate rebooted a host into Safe Mode to blind EDR. Defender named the ransomware and could not remove it for 28 minutes

Huntress documents the first Akira use of Safe Mode boot. The full technique chain contains no exploit at all, the encryptor starved of virtual memory and failed, and the victim is still extortable.

A single brass key resting on a stack of dark rectangular plates, macro close-up on a near-black surface, lit cyan along one side of the stack and crimson along the other.

Threat Intel14 Aug 2026

Jewelbug ran espionage and crypto fraud from one panel, with 580,000 stolen cookies logged. Symantec's own attribution doubts it was one team

One million implant check-ins, more than 15 government webmail tenants watering-holed by a single script tag, and a supplier relationship the coverage flattened into one team.

A square fingerprint reader sensor set into the brushed dark deck of a laptop, macro close-up beside the corner of a trackpad, ringed in crimson light against near-black.

Threat Intel13 Aug 2026

The Briefing

Two Tanium report permissions can become code execution. Three release branches need checking.

Two perfect scores in this Patch Tuesday need no action. The two being exploited score 7.8.

Adobe told Commerce merchants to rotate their payment gateway keys. The coverage stopped at the patch.

A national CERT found six RouterOS flaws with a cyber model, and wrote down exactly what the model did not do

SonicWall's exploited SMA 1000 flaws need four remediation steps, and almost everyone reported one

Cisco found these with frontier AI models, grouped them into seven umbrella CVEs, and never said how many bugs there are

The KEV remediation window fell from 21 days to three, and the 60-day tier everyone quotes has never existed

A BGP hijack delivered a malicious Virtualizor update, and the TLS certificate on it was genuine

The whole public evidence base for this morning's Falcon zero-day is 744 characters, and two of its sentences did not survive the retelling.

CISA catalogued three exploited flaws this evening that nobody has reported. Two of them cannot be found by vendor name.

A Chinese-speaking crew turned Brazilian .gov sites into SEO infrastructure. Their Apache module strips the CSP on three paths and leaves the rest correct.

SonicWall, JFrog and Langflow were reported exploited on one source each. CISA catalogued two of them eight hours later.

Packagist removed thirteen malicious Composer themes. A fourteenth from the same operator is still listed.

A fake coding test forbade AI assistants. The researchers say an AI would have caught the malware.

A 9.8 JFrog Artifactory bypass was exploited four days after disclosure. The patch shipped on day one.

UAC-0099 hid a nuclear weapons essay in malware to stop AI analysis. Nobody has tested whether it works.

DoJ corrected its China hacking claim. It fixed one copy of two, and 17 articles still carry the original

TerminalFix chains eight stages from a fake CAPTCHA to a reverse tunnel, and never touches a CVE. There is nothing to patch

Four CVEs let agent tools run without the model running at all. The guardrails were not bypassed, they were never asked

An Akira affiliate rebooted a host into Safe Mode to blind EDR. Defender named the ransomware and could not remove it for 28 minutes

Jewelbug ran espionage and crypto fraud from one panel, with 580,000 stolen cookies logged. Symantec's own attribution doubts it was one team

Unit 42 extracted every synced passkey from a Windows machine. Ars Technica called it a nothingburger

The August zero-day was North Korea. The part still working is the job advert

Microsoft patches 420 flaws and 3 zero-days. Not one zero-day is rated Critical

Credentials did not stop mattering. The DBIR headline was read wrong