CTIPilot
AI-generated · no human review · verify critical claims against the linked source. how it works →
‹Sun · 15 Nov 2026
All daily briefs →
Daily brief · UTC day

Sunday, 15 November 2026

4 verified findings from 2 runs · 16 updates to prior coverage · the settled record for this UTC day, in the classic brief order.

Criticality
Kind
Topic
Region
TL;DR · the day in one read
  1. 01ARTEX in AWS: CloudWatch log searches for secrets, Cognito self-sign-up for credentials; hunt CloudTrail for both. Datadog Security Labs reports that in October 2026 operators of the ARTEX agentic penetration-testing framework worked against several AWS environments with a focus on financial data: sessions opened with anonymous S3 reads, a stolen long-term IAM key drove CloudWatch Logs FilterLogEvents searches for company names, passwords and access keys, and where no key was held the agent self-registered in a public Cognito user pool and traded the tokens for AWS credentials. Datadog says the distributions it studied do not appear to carry a cloud playbook of their own, so the steps follow the model behind them. →
  2. 02Ransom-ISAC: Gentlemen affiliate, VPN without MFA to Domain Admin in an hour; GPO locker likely lost a replication race. Ransom-ISAC's incident-response report describes an early-August 2026 intrusion by a Gentlemen ransomware affiliate that began with an SSL-VPN account without MFA and reached Domain Admin in under an hour, most likely with credentials taken from an end-of-life Windows Server 2003 host that a dormant local administrator account opened, then used a Rust credential harvester and the EtherRAT implant. The locker's GPO disabled antivirus across the estate but the task that runs it most likely never reached clients because of a replication race, so the locker ran on one host; the report names the telemetry at each step. Single source, no victim named. →
  3. 03Huntress: 2,223 crypto-named fake sessions sit on 118 ScreenConnect servers, likely bait for unpatched technicians. Huntress counted 2,223 unauthorized guest sessions on 118 self-hosted ScreenConnect servers at 109 MSP and organisation accounts, enrolled between 2026-11-03 and 2026-11-13 and named after crypto exchanges; on the 29 servers with complete history an operator revisited every 151 seconds to keep them online. Huntress's leading hypothesis, which it has not observed, is that they are bait for a technician with a client older than 26.6.5 to connect, so that CVE-2026-84869 runs a file on the technician's workstation; it saw no malicious activity from the sessions and found no overlap with the SCKit indicators. →

01Active threats, incidents & disclosures4 items

NOTABLENATOB2

ARTEX against AWS: Datadog sees agent sessions open with anonymous S3 reads, search CloudWatch Logs for secrets with stolen long-term IAM keys, and self-register in public Cognito user pools to obtain AWS credentials

Datadog Security Labs reports that in October 2026 operators of ARTEX, the agentic penetration-testing framework Microsoft ties to Storm-0940, worked against several AWS environments with a focus on financial data, and that every session it reviewed opened with unauthenticated S3 reads, which CloudTrail records with a userIdentity type of AWSAccount and an accountId of anonymous, calling GetObject, ListObjects and similar read APIs (Datadog Security Labs, 2026-11-09). Where S3 hosted a web front end, the agent read build manifests and runtime configuration, and it requested objects named like placeholders just before bursts of key guesses (Datadog Security Labs, 2026-11-09), the habit Wiz saw as a request for a path that should not exist before each fuzzing run (Wiz Research, 2026-10-22).

With a valid long-term IAM access key, the agent's most frequent call was CloudWatch Logs FilterLogEvents, always with a limit of 1,000 events: it first searched log text for the names of cryptocurrency firms, hedge funds and invoicing terms and, in a later session on 2026-10-20, for storage links, internal addresses, database references, passwords and AKIA-style access keys (Datadog Security Labs, 2026-11-09).

Without a key, the agent self-registered in a public Cognito user pool, iterating against a PreSignUp Lambda that checked a tenant domain, then ran InitiateAuth, RespondToAuthChallenge, GetUser and GetTokensFromRefreshToken and traded the tokens for short-term AWS credentials, which CloudTrail shows under the CognitoIdentityCredentials role session name; some SignUp calls named the tool in the user agent (Datadog Security Labs, 2026-11-09). It then tried read-only discovery calls such as ListBuckets, ListSecrets, DescribeParameters, ListTopics, DescribeLogGroups, GetRestApis, DescribeRepositories and ListFunctions, all of which failed because the identity pool's authenticated role was narrow (Datadog Security Labs, 2026-11-09).

Datadog found hundreds of internet-exposed ARTEX consoles, often in Mandarin, and says the distributions it analysed do not appear to carry a cloud methodology of their own, so the post-exploitation playbook depends on the model in use; some versions block .gov, .gov.cn, .edu and .edu.cn targets by default (Datadog Security Labs, 2026-11-09).

Triage: FilterLogEvents is a routine call for log tooling and engineers; the discriminator is a long-term user key searching log text for credentials, internal addresses or company names at a constant page limit, from an address that key does not normally use.

In October 2026, Datadog observed in-the-wild activity from ARTEX, an AI-native penetration framework, against several AWS environments, with a focus on accessing financial data.

Before either, every session we reviewed opened with unauthenticated S3 requests.

We saw the agent go further, registering an account, collecting tokens, trading them for AWS credentials, then enumerating resources.

In this case, all of them failed because the identity pool's authenticated role did not grant the required permissions.

Datadog Security Labs 2026-11-09

Builds on: FBI: various actors run ARTEX against many sectors, government included; Microsoft names… · Elastic: CyberStrikeAI against AWS, six medium or low alerts joined only by source IP; alert on… · Entra app-credential abuse: agents hunt GitHub for secrets, an operator reaches MSP clients… · WorkOS breach agent was a researcher's ARTEX test that escaped its sandbox; legacy API, dev…

threat15 Nov 15:09Zsingle-sourceOpen finding →
NOTABLENATOC2

The Gentlemen affiliate: an SSL-VPN account without MFA, a Windows Server 2003 host and a dormant local administrator gave Domain Admin in under an hour; the GPO locker most likely lost a replication race, so antivirus was disabled estate-wide but only one host was encrypted (Ransom-ISAC)

Ransom-ISAC reconstructs an early-August 2026 intrusion into a small company by a Gentlemen ransomware affiliate from forensic images of two domain controllers, event logs of 41 hosts and firewall logs (Ransom-ISAC, 2026-11-13). Entry was an SSL-VPN account without MFA that succeeded on the first attempt from an unusual country; neither egress address appears in any earlier failed login, so the credential was most likely held beforehand. Within the hour the actor ran an anonymous SMB sweep, three Kerberoast-style service-ticket requests (the KDC logged an RC4 and DES downgrade against an AES-only account) and hundreds of failed SMB logons against a domain controller, then used a dormant, weak-password local administrator on an end-of-life Windows Server 2003 SQL host that allowed null sessions. A Domain Admin session opened on the domain controller less than an hour after the first login; Ransom-ISAC could not prove how and says the credentials were most likely taken from that host. The actor then pushed a Rust credential harvester, which takes the DPAPI domain backup key, and EtherRAT, a Node.js implant that reads its command address from an Ethereum smart contract and never reached it; data left through the VPN tunnel over about two days of a weekend (Ransom-ISAC, 2026-11-13).

The locker, run on a domain controller in domain-wide mode, built a GPO linked enforced at the domain root and the Domain Controllers container, with eight registry values that disable antivirus and a Preferences scheduled task in SYSVOL. The call that declares the Preferences extension reached a domain controller the GPO had not replicated to, the script suppressed the error and no client read the task: antivirus was off estate-wide but the locker ran on one host, which wiped its logs (Ransom-ISAC, 2026-11-13).

Triage: scanners and inventory tools also sweep SMB anonymously and legacy applications use null sessions (the finding is a dormant account logging on right after); developers resolve Ethereum RPC providers, so scope that rule to servers; administrators edit GPOs, so correlate with the change calendar (Ransom-ISAC, 2026-11-13).

Everything here comes from a single incident response engagement.

The GPO still turned off the antivirus across the estate, but the locker ran on only one host.

Ransom-ISAC 2026-11-13
threat15 Nov 04:28Zsingle-sourceOpen finding →
Sources: Ransom-ISAC
NOTABLENATOB2

ScreenConnect: Huntress finds 2,223 fake guest sessions named after crypto exchanges on 118 self-hosted servers, enrolled from 2026-11-03 and, in its leading hypothesis, bait for technicians whose clients predate 26.6.5; it saw no malicious use and no overlap with SCKit

Huntress counted 2,223 fake guest sessions on 118 self-hosted ScreenConnect servers across 109 MSP and organisation accounts, enrolled between 2026-11-03 and 2026-11-13: always-online "computers" named after crypto exchanges (an exchange word joined to a role word) with a trading-desktop thumbnail (Huntress, 2026-11-14). An MSP asked about them on 2026-11-12; its own server showed no web shells, suspicious child processes or new accounts, and the first Access installer Huntress examined was publicly downloadable, which ScreenConnect allows without authentication by design, and a guest client needs the server's relay address, which is embedded in that server's guest installer, to register. Huntress saw no malicious activity from the sessions: technicians who opened them mostly stayed under 30 seconds, and in one case a technician used the send-text feature, which ScreenConnect does not log, so Huntress treated it as a possible credential exposure. On 29 servers with complete history the operator revisited every 151 seconds to keep the sessions online (Huntress, 2026-11-14).

Huntress's leading hypothesis, which it has not observed, is bait: a technician who connects with a client older than 26.6.5 would let a malicious guest use the file-transfer flaw CVE-2026-84869 to run a file on the technician's workstation, often the MSP's most privileged machine; it cannot rule out capturing what a technician types, and it found no overlap with the published SCKit indicators (Huntress, 2026-11-14). It considers the public suggestion that the sessions are a malware sandbox detonating installers unlikely.

Triage: a malware-analysis sandbox detonating a ScreenConnect installer also creates guest sessions; per Huntress, a sandbox session sends a key and JPEG screenshots of the analysis VM, while these send no key, reuse seven screenshot families and randomise hardware and OS details (Huntress, 2026-11-14).

As of this writing, we've found 2,223 fake sessions on 118 servers across 109 MSP and organization accounts.

This is our leading hypothesis, but it's important to note that we have not observed the guest sessions being used this way.

Huntress 2026-11-14

Builds on: ConnectWise patches a ScreenConnect flaw Huntress had already watched turn every infected… · ScreenConnect SCKit: second wave hits on-prem; disable RMM+, update technician clients to…

threat15 Nov 04:27Zsingle-sourceOpen finding →
Sources: Huntress
NOTABLENATOB2

PolinRider: a North Korea-linked loader reaches developers through force-pushed forged commits in at least 7,609 GitHub repositories, hijacked npm releases and a fake employer's take-home test, and reads its server address from an Ethereum wallet

Apiiro calls PolinRider a North Korea-linked worm and found it in at least 7,609 GitHub repositories owned by 2,330 accounts, mostly developers' own (Apiiro, 2026-11-09). Three routes, none relying on an npm install script. Forged commits: with stolen GitHub access the worm force-pushes a copy of a commit the owner already made, with the same author, message and date, plus a loader hidden after hundreds of spaces in a build configuration file; in 94% of 9,990 injection pushes the copy claims to be more than an hour older than the push, so git log shows an old commit under the owner's name. Cloning runs nothing; building the project or running its dev, build or lint scripts loads the file, and in the newest wave opening the folder in VS Code is enough (Apiiro, 2026-11-09). Hijacked npm releases: Sonatype found concealed loaders in 12 releases across 10 hijacked packages with no install or postinstall hook, so disabling install scripts does not help (Sonatype, 2026-11-04). A fake employer: JFrog traced AccuraSoftHire, a decoy recruiter with ID-verified LinkedIn recruiters and job ads, whose take-home loads a look-alike npm package and whose scripted camera check fails and hands the candidate a command that installs the same loader (JFrog, 2026-11-09); StepSecurity says taking its site down, as had happened by 2026-11-13, does not end the operation (StepSecurity, 2026-11-13).

The loader reads the latest outbound transaction of a hard-coded Ethereum wallet and decodes two IPv4 addresses from its recipient field (Apiiro, 2026-11-09), then fetches XOR-encoded JavaScript over plain HTTP on port 443, so blocking one server changes nothing; the wallet in the hijacked npm releases also appears in GitGuardian's GhostAction report, which Sonatype says does not establish a shared operator (Sonatype, 2026-11-04). Cleanup does not hold: 1,949 repositories were infected again after being publicly listed, and where owners cleaned up the stolen access often still worked; Apiiro counts roughly 2,300 developers reached (Apiiro, 2026-11-09).

Triage: blockchain developers legitimately call public RPC endpoints; the discriminator is a build or lint process in a project with no Web3 purpose doing so and then spawning a detached node child (Apiiro, 2026-11-09).

Apiiro research found the worm, PolinRider, in at least 7,609 GitHub repos owned by 2,330 accounts.

Cloning runs nothing.

Apiiro 2026-11-09

Disabling npm install scripts does not address code that runs during module loading or application startup.

Sonatype Research 2026-11-04

We have found no link between AccuraSoftHire and that intrusion.

JFrog Security Research 2026-11-09

Builds on: GhostAction: fake security workflows now harvest whole git histories; rotating Actions secrets…

threat15 Nov 04:26Zmulti-sourceOpen finding →

02Updates to prior coverage16 items

HIGHCVE-2026-109618 +2exploitedupdatedNATOA2

CVE-2026-109618 / CVE-2026-109619 and CVE-2026-123039, Check Point Force, Spark and Cloud firewalls: PHP2Shell on the Mobile Access and VPN web interfaces, chained with a local root escalation, is exploited against unpatched gateways (implant, ESXi wiper); the fix is in Jumbo takes R81.20 take 174, R82 take 132, R82.10 take 52 and R82.20 take 14

First published 2026-10-29 · open finding →

Improvementrun 2026-11-15T1407Z-auditsourcesbody

Check Point's hotfix article says R82.20 gateways are protected by the built-in "PHP Remote Code Execution" IPS signatures even with IPS disabled and HTTPS Inspection inactive, and Check Point Research saw hundreds of segmentation faults in the portal's Apache and PHP workers during the exploitation window, a crash pattern defenders can look for.

Check Point's hotfix article states that R82.20 firewalls protect themselves through the "PHP Remote Code Execution" IPS signatures even when IPS is not enabled and HTTPS Inspection is not active, while the fix itself ships in the R82.20 Jumbo Hotfix Accumulator from take 14 (Check Point, 2026-11-12). Check Point Research reports that during the exploitation window the Mobile Access portal's Apache and PHP workers recorded hundreds of segmentation faults, which it reads as evidence of repeated exploitation attempts; a burst of portal worker crashes on a gateway with Mobile Access enabled is therefore a signal to investigate (Check Point Research, 2026-11-12).

NOTABLECVE-2026-91403 +3updatedNATOA1

eMudhra: the private keys of eight ECDSA intermediate CA certificates were trivially recoverable, so anyone could forge trusted HTTPS certificates, and eight RSA intermediates were revoked on 2026-11-07 for key handling eMudhra cannot show compliant; browsers block the eight ECDSA intermediates, clients with their own trust store or no revocation checking need an update (CVE-2026-111943 in Go)

First published 2026-10-21 · open finding →

Improvementrun 2026-11-15T1407Z-audittitlesummarysourcessourcing_noteactionsbody

Microsoft's post now carries an update dated 2026-11-12 whose table adds VS Code 1.147.0 (2026-11-18), Azure CLI 2.93.0 (2026-11-16), which removes two more eMudhra roots, the .NET SDK update in December 2026 and the API Management self-hosted gateway and Microsoft Build of OpenJDK after the Mozilla release of 2026-11-16 to the products that drop the eMudhra roots.

Microsoft's post now carries an update dated 2026-11-12 with a per-product table: Visual Studio Code 1.147.0 is due with fixes on 2026-11-18 after the node-vsce-sign package of 2026-11-12, Azure CLI 2.93.0 is due on 2026-11-16 to remove two more eMudhra roots, emSign Root CA - C1 and emSign ECC Root CA - C3, from its bundled trust store after 2.92.0 blocked the G3 root on 2026-11-03, the .NET SDK follows in December 2026, and the API Management self-hosted gateway and the Microsoft Build of OpenJDK follow the Mozilla release of 2026-11-16 (Microsoft, 2026-11-12).

Sources: Google (Chrome Root Program and Android Security) · Dennis Jackson (eMudhra intermediate CA compromise incident tracker) · Mozilla Security Blog · Go team (golang-announce) · BleepingComputer · Oracle (Security Alert CVE-2026-91403) · OpenJDK (Vulnerability Advisory 2026/10/21) · NCSC.ch Security Hub (advisory 13091) · Oracle (Security Alert CVE-2026-114820) · OpenJDK (Vulnerability Advisory 2026/10/24) · OpenJDK (Compromised Certificates and OpenJDK Java Runtimes) · Python Software Foundation (security-announce, CVE-2026-115557) · ENISA EUVD (EUVD-2026-109299) · Cisco PSIRT (cisco-sa-emudhra-info, version 1.1) · Python Discourse (eMudhra information for Python users) · The Hacker News · eMudhra (press release) · Microsoft Security Response Center (updated 2026-10-27) · Cisco PSIRT (cisco-sa-emudhra-info, version 1.5) · GitLab (patch release 19.5.3) · Python Insider (release post) · Dennis Jackson (eMudhra intermediate CA compromise incident tracker, update of 28 October) · Oracle (Security Alert CVE-2026-114820, revision 2) · Cisco PSIRT (cisco-sa-emudhra-info, version 1.8) · Google (Chrome Root Program and Android Security, update of 2026-11-04) · Apple Support (information on the compromised eMudhra intermediate CAs, translated from German) · heise online (translated from German) · SonicWall (informational advisory SNWLID-2026-0021) · SonicWall PSIRT (advisory data for SNWLID-2026-0021) · Apple Support (information on the compromised eMudhra intermediate CAs, revised 2026-11-10, translated from German) · Microsoft Security Response Center (updated 2026-11-10) · Microsoft Security Response Center (updated 2026-11-12) · Cisco PSIRT (cisco-sa-emudhra-info, version 1.11) · eMudhra (Mozilla CA Program bug 2083449, full incident report) · eMudhra (Mozilla CA Program bug 2083449, comment 11)
NOTABLECVE-2026-111497exploitedupdatedNATOA2

CVE-2026-111497, Brotli decoder flaw in Siemens industrial and software products (Desigo CC, SIMATIC WinCC, Industrial Edge, Teamcenter and Opcenter among them), Mitel MiVoice Business and OpenScape, with Cisco listing Duo Network Gateway, RoomOS and Nexus 3000 and 9000 switches (Guest Shell) in its combined advisory: Desigo CC takes the .NET updates, while most Siemens products, the Mitel updates and the Industrial Edge Common Configurator fix are still pending

First published 2026-11-07 · open finding →

Updaterun 2026-11-15T1407Z-audittitlesummaryaffected_productscvessourcesbody

Cisco's combined advisory, at version 1.13 of 2026-11-13, now lists Nexus 3000 and 9000 switches running standalone NX-OS with Guest Shell as affected, fixed by Guestshell 4.9 or NX-OS releases due from December 2026 to February 2027, and Apple fixed the flaw in tvOS 27.2.1 on 2026-11-13.

Cisco's combined advisory, revised to version 1.13 on 2026-11-13, lists Nexus 3000 and Nexus 9000 Series switches in standalone NX-OS with Guest Shell as affected under bug CSCwx56979, without tying the rows to one of the four Node.js CVEs the advisory covers, fixed by Guestshell 4.9 or by NX-OS 10.3.11 (December 2026), 9.3.19, 10.5.7 and 10.7.1 (January 2027) or 10.4.9 and 10.6.5 (February 2027) (Cisco PSIRT, 2026-10-27). Apple fixed the flaw in tvOS 27.2.1 on 2026-11-13 for Apple TV 4K 2nd generation and later, with the same Brotli entry as its other notes (Apple, 2026-11-13).

NOTABLEupdatedNATOB2

LLMjacking: stolen cloud identities and keys enable managed model services in victim accounts and resell the quota through proxy gateways (GTIG and Mandiant), with control-plane hunting guidance for AWS, Azure and Google Cloud

First published 2026-11-09 · open finding →

Improvementrun 2026-11-15T1407Z-auditreferencessourcesbody

GTIG attaches CVE-2026-40217 to the LiteLLM /key/regenerate user-ID rebind it saw exploited, but Obsidian Security tracks CVE-2026-40217 as a sandbox escape in LiteLLM's custom-code guardrail, with the fixes for its reported chain in v1.83.14-stable.

GTIG names the /key/regenerate user-ID rebind CVE-2026-40217 and advises patching model gateways against it (GTIG, 2026-11-02). Obsidian Security tracks CVE-2026-40217 as a sandbox escape in LiteLLM's custom-code guardrail, which X41 D-Sec separately showed through the guardrail test path, not a key-regeneration flaw, and says the full set of fixes for the chain it reported shipped in LiteLLM v1.83.14-stable on 2026-04-25 (Obsidian Security, 2026-09-08). The id therefore does not match the flaw GTIG describes; LiteLLM v1.83.14-stable or later is the release that carries the fixes for the disclosed chain, and no source names a separate fix for the rebind.

NOTABLEupdatedNATOB2

BluePhishing: Proofpoint's proof of concept turns a consent link for a third-party Entra agent blueprint into tenant initial access, with an agent-creation permission that Proofpoint says cannot be revoked

First published 2026-11-10 · open finding →

Improvementrun 2026-11-15T1407Z-auditheadlinesummarysourcesbody

Microsoft's message MC1506733 schedules the change that lets default Entra users create agent blueprints from 2026-11-04, with the rollout due to finish by late November, unless an administrator turns off the "Users can create agent blueprints" setting, which is the lever a tenant has now.

Microsoft's Message Center item MC1506733 says that starting 2026-11-04 default users will be able to create new agent blueprints unless an administrator turns off "Users can create agent blueprints", with the worldwide rollout beginning in early November 2026 and expected to complete by late November 2026; the setting sits under Entra ID user settings, Default user role permissions, and users assigned the Agent ID Developer role keep the right regardless (Microsoft 365 Message Center, 2026-10-13).

NOTABLEupdatedNATOB2

GhostFrame: Recorded Future's Insikt Group finds an ESXi post-exploitation framework (vmhostd-core implant, VM control panel, VMkatz credential theft) used by TAG-218, whose activity overlaps Settra ransomware and which targeted SonicWall, Palo Alto Networks and Fortinet VPN appliances in the US and Europe

First published 2026-11-11 · open finding →

Correctionrun 2026-11-15T1407Z-auditbody

Insikt gives the persisted file as /opt/vmhostd-core and vmhostd as the process name the implant presents, a name close to the genuine vmware-hostd daemon, and hedges its VMkatz reading: an operator may be able to obtain guest credentials without deploying malware inside each VM.

Insikt Group gives the persistence file as /opt/vmhostd-core and says the implant presents its process as vmhostd, a name resembling the genuine vmware-hostd daemon; either the file or the process is the indicator it lists (Recorded Future, 2026-11-10). Its reading of the VMkatz capability is hedged: the capabilities suggest that an operator controlling an ESXi host may be able to obtain credentials from hosted Windows systems without first deploying credential-stealing malware inside each VM (Recorded Future, 2026-11-10).

NOTABLEupdatedNATOB2

CopperDragon: Check Point ties three China-nexus intrusion sets to one Sparrow-framework supplier, among them SilentDelivery watering holes on government, shipping, IT and research sites that show a fake Chrome certificate warning (Czech Republic, Germany and France in its telemetry)

First published 2026-11-11 · open finding →

Correctionrun 2026-11-15T1407Z-auditbody

Check Point says the operators placed their tools in existing system directories and renamed SoftEther VPN to system-process names such as vmtoolsd.exe and csrss.exe, so the name, not the path, is the discriminator.

The SoftEther discriminator is the name, not the path: Check Point says the operators renamed SoftEther VPN to names such as vmtoolsd.exe and csrss.exe and deliberately placed their tools in existing system directories instead of creating new ones (Check Point Research, 2026-11-10), so a SoftEther binary or configuration under a system-process or VMware Tools name is the signal, wherever it sits.

NOTABLECVE-2026-98432updatedNATOB2

CVE-2026-98432, Linux kernel IPv4 TCP: a use-after-free in connection setup, present since 4.4, has an Anthropic-built remote root exploit against MPTCP listeners that CloudLinux could not reproduce; Citrix XenServer, XCP-ng, MikroTik, Red Hat, Debian, CIQ, Amazon Linux, Ubuntu's generic kernels for 22.04 to 26.10 and Sophos have fixes, Ubuntu 20.04 and older and some HWE, FIPS, NVIDIA and Azure variant kernels are in progress and WatchGuard and Fortinet have none for their products (CVSS 9.8)

First published 2026-10-31 · open finding →

Updaterun 2026-11-15T1407Z-auditaffected_productscvessourcesactionsbody

Sophos's advisory, now at version 5 of 2026-11-12, adds SD-RED firmware 3.0.012 as a manual patch for Sophos UTM (2026-11-10) and says end-of-life RED models get no fix; Ubuntu's tracker of 2026-11-14 lists most HWE, Azure, NVIDIA and Raspberry Pi variant kernels as fixed; Veeam's KB4966 of 2026-11-13 adds Recovery Orchestrator 13.1 and gives the delivery route per component; and Debian's current trixie security build is 6.12.117-1.

Sophos's advisory, updated to version 5 on 2026-11-12, adds that SD-RED firmware 3.0.012 was released on 2026-11-10 as a manually installed patch for Sophos UTM, which requires UTM 9.724 or later, and that legacy RED models past end of life are affected but will not receive a fix (Sophos, 2026-11-06). Ubuntu's tracker, updated on 2026-11-14, now lists the HWE kernels for 20.04, 22.04 and 24.04, Azure for 22.04 to 26.04, NVIDIA for 22.04 to 26.04 and Raspberry Pi for 22.04 to 26.04 as fixed, while FIPS for 24.04 and 20.04, Azure 7.0 and Azure NVIDIA for 24.04, NVIDIA 6.8 for 22.04 and the generic kernels for 20.04 and older remain in progress (Canonical, 2026-11-10). Veeam's KB4966, last modified on 2026-11-13, adds the helper appliances embedded in Veeam Recovery Orchestrator 13.1, fixed by a private fix, and Recovery Media for Veeam Agent for Linux 6.3, and lists the delivery per component: Veeam Updater for the appliances, an automatic update on launch for workers, a private fix for the helper appliances of Windows-based Backup & Replication and Recovery Orchestrator, and new recovery media (Veeam, 2026-11-10). Debian's tracker lists 6.12.117-1 as the current trixie security build, after the first fixed 6.12.113-1 (Debian Security Tracker, 2026-11-03).

HIGHCVE-2026-105133 +1exploitedupdatedNATOB1

CVE-2026-105133 / CVE-2026-105134, AhsayCBS backup management console: an unauthenticated authentication-bypass and code-execution chain exploited since 7 October to drop webshells and a cryptominer; Ahsay's earlier patches were incomplete, and its standard releases v9.17.0 and v10.3.6 now list a patchset for both flaws

First published 2026-10-10 · open finding →

Updaterun 2026-11-15T1407Z-audittitleheadlinesummarycvessourcesactionsbodysourcing_note

Ahsay's standard releases v9.17.0 (2026-11-12) and v10.3.6 (2026-11-02) list its patchset for both exploited CVEs and turn remote API access and the JSP engine off by default, so the fix no longer depends on the partner-only hotfixes, whose latest builds are v9.15.2.78 and v10.3.4.55 (2026-10-27); the current v10 build, v10.3.6.14 of 2026-11-13, also stops the Redirector logging the CBS password in readable form. CISA's catalog calls CVE-2026-105134 command injection, while Horizon3's source analysis shows a path-traversal file write.

Ahsay released AhsayCBS v9.17.0 on 2026-11-12 as a standard release; its release notes list a security patchset for CVE-2026-105133 and CVE-2026-105134, remote API access disabled by default and the JSP engine disabled to prevent execution of malicious JSP files (Ahsay, 2026-11-12). For v10, the release notes of v10.3.6, dated 2026-11-02, list the same patchset and the same two settings (Ahsay, 2026-11-02), and Ahsay announced v10.3.6.1 on 2026-11-03 (Ahsay, 2026-11-03). Partners who stay on hotfixes have newer builds: Ahsay's advisory of 2026-10-27 tells partners on v7, v8 and v9 to take the v9.15.2.78 hotfix, which adds security hardening, and names v10.3.4.55 as the latest v10 hotfix (Ahsay, 2026-10-27). The current v10 build, v10.3.6.14 of 2026-11-13, lists among its security fixes that the Redirector logged the CBS password in readable form in its access log, so access logs from earlier builds can hold that password (Ahsay, 2026-11-13). The flaw class of CVE-2026-105134 is described two ways: CISA's catalog calls it command injection through the random parameter of the UpdateReceivers.do endpoint (CISA, 2026-10-14), while Horizon3's analysis of the source describes a file-write path traversal that leads to code execution (Horizon3.ai, 2026-10-14).

HIGHupdatedNATOA2

PK Softech, software supplier of Swiss pension funds: malware and a data outflow at Publica, with the City of Zurich fund PKZH saying its insured persons' test data was stolen, the Bernese funds BPK and BLVK and Pensionskasse Post also reporting the supplier incident, and the Federal Prosecutor's Office investigating

First published 2026-10-09 · open finding →

Correctionrun 2026-11-15T1407Z-auditsourcesevidencebody

The Bernische Pensionskasse's notice PDF no longer resolves, so its citations now point to the fund's notice page, which carries the 9 October notice; that page does not say the supplier is the same company that serves Publica, so that fact is now cited to AWP, which reports it. A Keystone-SDA quote's German original now reads Darknet as the page does, a superseded gloss on the fund's earlier statement was dropped, and the section notes how two statements on fraudulent contacts differ in scope.

That the supplier is the same company that counts Publica among its customers is AWP's report (cash.ch (AWP), 2026-10-09); the Bernische Pensionskasse's notice of 9 October says that an external software supplier of the fund detected a cyber incident at the end of September 2026 but does not name its other customers (BPK, 2026-10-30). Two statements about contacts with insured persons differ in scope: Inside IT relayed on 2026-10-20 a report that the State Secretariat for Security Policy had seen first fraudulent contact attempts connected with Publica (Inside IT, 2026-10-20), while Publica says it knows of no case of a person demonstrably contacted or harmed in connection with the attack (Publica, 2026-11-10); neither names a case.

HIGHCVE-2026-126147 +18updatedNATOA2

Fortinet coordinated security hardening release: the advisories published on 2026-11-10 carry unauthenticated FortiMail, FortiAnalyzer, FortiManager, FortiOS and FortiAuthenticator flaws rated CVSS up to 9.8; FortiAnalyzer and FortiManager 8.0 have no fixed build for the JSON API flaw, FortiOS 7.0 and FortiMail 7.2 are left out of the October fixes

First published 2026-10-23 · open finding →

Correctionrun 2026-11-15T1407Z-auditsummarycvessourcesevidencesourcing_noteactionsbody

FortiAnalyzer and FortiManager 7.2 are fixed for CVE-2026-126229 in 7.2.13 and only 7.0 must migrate, so 7.2 owners update rather than migrate; the 7.6, 7.4 and 7.2 lines below their fixed builds are affected as well as 8.0. Fortinet added a virtual patch for the iBGP flaw CVE-2026-126254 on 2026-11-13, and FortiMail and FortiAuthenticator are dated to their 2026-10-21 release.

FortiAnalyzer and FortiManager 7.2 do not need to migrate for CVE-2026-126229: Fortinet's advisory lists 7.2.0 through 7.2.12 as affected and fixes them in 7.2.13, and only 7.0 must migrate; the 7.6, 7.4 and 7.2 lines below 7.6.8, 7.4.12 and 7.2.13 are affected as well as 8.0 (Fortinet PSIRT, 2026-11-10). Fortinet revised its advisory for the FortiOS iBGP flaw CVE-2026-126254, a numeric truncation that Fortinet rates as denial of service, on 2026-11-13 to add the virtual patch FG-VD-62507.0day in FMWP database update 26.118 (Fortinet PSIRT, 2026-11-10). Fortinet's blog dates the FortiMail and FortiAuthenticator builds to 2026-10-21 and the FortiOS, FortiManager and FortiAnalyzer builds to 2026-10-22 (Fortinet, 2026-10-21).

NOTABLECVE-2026-123158exploitedupdatedNATOA2

CVE-2026-123158, OpenSSH Portable before 10.6p2 (*sshole): on Linux with musl libc and no PAM, a password of 257 to 1,024 bytes logs in to accounts whose password field is *; NCSC Switzerland issued an advisory (CVSS 3.1 8.1)

First published 2026-11-06 · open finding →

Correctionrun 2026-11-15T1407Z-auditsummarysourcesbody

NCSC Switzerland has rated the flaw actively exploited, with a public proof of concept, since 2026-11-10 09:59, and QNAP's QSA-26-21 of 2026-11-13 says QTS, QuTS hero and QuTScloud are not affected.

NCSC Switzerland updated its advisory on 2026-11-10 at 09:59 to rate the exploitation status "Actively Exploited, PoC Available", replacing the unknown rating it carried until that morning (NCSC Switzerland, 2026-11-05). QNAP's advisory QSA-26-21 of 2026-11-13 states that its system library and account-authentication configurations do not meet the conditions for exploitation, so QTS, QuTS hero and QuTScloud are not affected (QNAP, 2026-11-13).

HIGHCVE-2026-124007 +7exploitedupdatedNATOB2

Dixell/Copeland XWEB EVO refrigeration supervisors: VulnCheck published 14 CVEs on 2026-11-12, seven scored CVSS 4.0 9.2 to 9.3 (unauthenticated command injection that runs as root, a hard-coded Administrator password, Administrator and root passwords computed from the MAC address and date); VulnCheck sees two of the 14 exploited, the injection flaws are fixed on the current branch and the password flaws will stay unfixed after Copeland ended support on 2026-03-16

First published 2026-11-13 · open finding →

Updaterun 2026-11-15T1407Z-audittitleheadlinesummaryprioritytagscvessourcesevidencesourcing_noteactionsbody

VulnCheck reports two XWEB EVO flaws exploited by opportunistic scanners: the unauthenticated configuration write CVE-2026-124014 (CVSS 4.0 8.8) from 2026-11-12 and root-command probes against CVE-2026-124010 from 2026-11-13. It says the injection and backdoor flaws are fixed on the current firmware branch, so the unfixed exposure is the pre-2021 fleet plus the predictable-password flaws; its blog and its CNA record disagree on whether CVE-2026-124014 reaches 5.8.8.

VulnCheck's canary sensors recorded in-the-wild exploitation of CVE-2026-124014, an unauthenticated write of the system configuration file through config.lua that affects XWEB EVO 4.3.0 through 5.8.8 (CVSS 4.0 8.8), from 2026-11-12 22:00 UTC, and four probes running a root command through CVE-2026-124010 from 2026-11-13 03:41 UTC; it added both to its own exploited-vulnerabilities list and reads both sources as opportunistic scanners, not the October refrigeration campaign (VulnCheck, 2026-11-13; VulnCheck, 2026-11-12). The same disclosure says the injection and backdoor-password class was removed in the 2021 rewrite and is fixed on the current branch, so the unfixed exposure is the pre-2021 fleet plus the predictable-password flaws, which reach the last firmware (VulnCheck, 2026-11-13). For CVE-2026-124014 the two VulnCheck sources differ: the blog lists it with the pre-2021 class, while the CNA record gives XWEB EVO 4.3.0 through 5.8.8 as affected (VulnCheck, 2026-11-12). Copeland's guide tells operators to install the latest build for their model from its System Software Update page while they migrate (translated from French) (Dixell France, 2026-10-14).

NOTABLECVE-2026-119618updatedNATOA2

CVE-2026-119618, Apache HTTP Server 2.4.55 to 2.4.70: a race in mod_http2 assign_headers() corrupts heap memory and leads to remote code execution on servers with HTTP/2 enabled; fixed in 2.4.71, Red Hat calls it unauthenticated and claims public exploit research, no exploitation reported

First published 2026-11-12 · open finding →

Updaterun 2026-11-15T1407Z-auditheadlinesummarycvessourcesactionsbody

SUSE fixed CVE-2026-119618 for SUSE Linux Enterprise 15 SP7 with apache2 2.4.71-150700.4.30.1 (SUSE-SU-2026:5231-1, 2026-11-13), and its CVE page lists fixed packages for its LTSS lines and openSUSE Tumbleweed while other products are pending; CISA ADP scores the flaw CVSS 3.1 8.1 (high attack complexity) against Red Hat's and SUSE's 9.8, and Debian's tracker now lists forky and sid as fixed in apache2 2.4.71-1.

SUSE released apache2 2.4.71 for SUSE Linux Enterprise 15 SP7 (package 2.4.71-150700.4.30.1) under SUSE-SU-2026:5231-1, rated critical (SUSE, 2026-11-13), and SUSE's CVE page now lists fixed apache2 packages for the SUSE Linux Enterprise 15 SP4 to SP6 and 12 SP5 LTSS lines and openSUSE Tumbleweed while other SUSE products are still in progress (SUSE, 2026-11-13). The scores differ by assessor: Red Hat and SUSE give CVSS 3.1 9.8 with low attack complexity, while CISA ADP gives 8.1 with high attack complexity (Red Hat Product Security, 2026-11-11; SUSE, 2026-11-13). Debian's tracker now lists forky and sid as fixed in apache2 2.4.71-1 (Debian Security Tracker, 2026-11-15).

HIGHCVE-2026-125350exploitedupdatedNATOA1

CVE-2026-125350, Google Chrome before 157.0.8103.50: a use-after-free in the XML component is exploited in the wild and listed by CISA; Edge 157.0.4341.46 and Edge for Android carry the fix, other Chromium browsers are still to follow

First published 2026-11-11 · open finding →

Updaterun 2026-11-15T1407Z-audittitleheadlinesummarycvessourcing_noteactionsbodysources

Microsoft's Edge security release notes now list Edge 157.0.4341.46 of 2026-11-09 as carrying the fix for the exploited XML flaw, so Edge desktop is no longer pending; Microsoft's Security Update Guide added Edge to the CVE's record on 2026-11-13 with CISA ADP's CVSS 3.1 score of 8.8.

Microsoft's Edge security release notes now carry a Stable-channel row for Edge 157.0.4341.46 of 2026-11-09 stating that "this update contains a fix" for CVE-2026-125350, which the Chromium team reported as exploited in the wild (Microsoft, 2026-11-11). Microsoft's Security Update Guide added Edge to the CVE's record on 2026-11-13 and carries CISA ADP's CVSS 3.1 score of 8.8 (Microsoft, 2026-11-13). Edge desktop is therefore patched from that build on; the open exposure is limited to Chromium-based browsers whose vendors have not yet listed the fix.

NOTABLEupdatedNATOB1

Cursor marketplace: a tampered Claude Code extension (versions 2.1.312 and 2.1.313) planted an SSH key, a reverse tunnel and persistence for about eleven hours on 2026-10-29; uninstalling it does not remove the persistence

First published 2026-10-31 · open finding →

Updaterun 2026-11-15T0356Z-intelsummarysourcesevidencetechniquesactionssourcing_notebody

Cursor's own report of 2026-11-13 gives the origin the entry lacked: an access key Cursor issued to its vendor Fireworks AI was exposed, the attacker copied a bucket of coding-agent traces holding some unredacted credentials, and a stolen credential let it write the extension onto the marketplace cache; Open VSX never carried it and the last attacker activity, on 2026-11-01, was blocked.

Cursor's incident report of 2026-11-13 says an attacker obtained an access key Cursor had issued to its vendor Fireworks, which Fireworks exposed, and used it to copy a read-only storage bucket of historical coding-agent traces from sessions not in Privacy Mode; a small number of the traces held credentials that had not been redacted, and less than 0.01% of customers had API credentials exposed (Cursor, 2026-11-13). From those credentials the attacker reached select SaaS tools and, through them, archived non-production source code repositories, limited customer support data and credentials users had sent to Cursor support (Cursor, 2026-11-13). On 2026-10-29, "using a stolen credential from the exposed data from our internal system", the attacker placed the malicious copies directly on the cache that serves the marketplace; Cursor says Anthropic did not publish them, that this was not a vulnerability in Anthropic's product, that Open VSX never had the malware, that at most 5,000 users who requested specific cached versions that day potentially received the build, and that the last attacker activity, on 2026-11-01, was blocked (Cursor, 2026-11-13). Where credentials were exposed Cursor emailed team admins a list of the keys to rotate (Cursor, 2026-11-13).

Fireworks' bulletin of 2026-11-09 describes its own intrusion, which it links to a financially motivated group that sells tokens from unauthorized access to language models to illicit routing gateways ("LLMjacking"), and further activity on 2026-10-23 and 2026-10-24 after a security incident at a cloud provider that hosted its development machines; it does not mention Cursor (Fireworks AI, 2026-11-09). Neither company names an actor for the extension.

03Action items17 items

Verification & coverage notes2 runs

2026-11-15T1407Z-audit · audit · Opus 5.5 · window 170 h · 1 entry published

Scope and outcome

Scheduled weekly audit of 2026-11-08T13:09Z to 2026-11-15T14:07Z (anchor: the 2026-11-08 audit, the last with re-sweeps). Fifty-six entries with activity from the week's seven intel fires were truth-checked whole in six passes: 27 clean, 24 with imprecisions, 5 with factual errors (Fortinet 7.2 fix, SSHole NCSC.ch rating, Chrome/Edge fix, Apache httpd SUSE fix, Dixell XWEB EVO exploitation and fixes). Twenty-seven entries carry one record each from this fire (6 updates, 10 corrections, 11 improvements; 12 internal), and one entry was recovered (Datadog's ARTEX activity in AWS). The report is docs/audits/2026-11-15-quality-audit.md.

Verification & coverage notes

  • Claims: the open partial claim of cleanup wave 05 (2026-10-21T1147Z-audit, partial 2026-10-21T11:47:42Z wave=05) still stands on main; the scheduled audit ignores a partial claim (Phase 0 step 2), and it is more than 48 h old.
  • Scope: 38 entries filed in the window and 18 older entries the week's fires updated; 23 entries set aside as checked by the 2026-11-08 audit's own loop (20 with only its records in the window, 3 it recovered).
  • Truth passes: six batches, 1,646 ledger claims, every claim answered. Pre-pass leads: the Danfoss French original carried a stray word (fixed); the out-of-scope LiteLLM quote sits in a collapsed GitHub comment and is now bound to Wiz, which carries it.
  • Declined truth-pass findings, with reasons in the report: Danfoss "France alone" (Unit 42's figure caption says France), seven advisory technique-description gaps at the word bound, the Qubole self-claim, three append-only or metadata advisories, Node2Shell release-page sources.
  • KEV: twelve additions since 2026-11-08 (work/2026-11-15T1407Z-audit/kev-window.txt), eight covered; GPUStack, Openwave, BroadWorks and UnrealIRCd carry the week's fires' borderline-drop lines. No COVERED-STALE or RANSOMWARE row.
  • Stale-truth sweep: exploitation-consistency passes over the 56 window entries; supersession defects found by the passes (Chrome headline, Apache action, Dixell title and summary, AhsayCBS headline and action) are fixed through their records.
  • Calibration alarm: high 14 of 56 (25.0%); all 14 re-graded; Certighost moved to notable, Dixell XWEB EVO moved to high.
  • Backlog: empty at Phase 0. Duplicates: no CVE-sharing pair touching the window lacks a references[] link.
  • Recovered entry: Datadog's 2026-11-09 post on ARTEX in AWS clears PD-11(d) (cloud control-plane detection for an agentic framework the store tracks), at notable; it references the ARTEX, CyberStrikeAI, exposed-app-credentials and WorkOS entries as distinct findings.
  • borderline-drop: OpenAM 16.1.4 (CVE-2026-123549/-123550/-123551), Azul Payara (CVE-2026-110003, CVE-2026-90908), Cisco BroadWorks CVE-2026-20320, GPUStack CVE-2026-102895, UnrealIRCd CVE-2026-114713, Node.js 2026-11-11 releases, React Server Components CVE-2026-126187, Adobe AEM APSB26-189, Roundcube 1.7.6, Drupal GraphQL Core Schema, Plex2Shell: the week's fires' drop grounds hold (no constituency footprint shown, authenticated, or regular patch cycle).
  • borderline-drop: Oracle's pre-announcement of the 2026-11-17 Critical Patch Update (the release falls to the 2026-11-17 and 11-18 fires); VulnCheck's Langflow overlap with the agentic secret-hunting address (one vendor canary, no CVE).
  • borderline-drop: Danfoss: Belgium's "indications" of Iranian links (one wire relay) and CERT-FR's naming of Shahid Shoostari for earlier attacks; Erowa, Gleis D/Medi Porta, Liechtenstein register investigation, Vaud CYBER26, Microsoft 365 outage-themed phishing, Operation Endgame 2026-11-12, Saxony pagers, Berlin forensic interview, France NIS2 bill, BACS ESTV warning: no vector, nexus or obligation change.
  • borderline-drop: Huntress ClickFix roundup, Malbear Operation Snowball, Proofpoint Europe's Threat Landscape 2026, Kaspersky Q3 statistics, Sophos retail ransomware, Krebs on the Qilin administrator, Morphisec, TrendAI MCP and Hermes, Enea SilentAir, Greeter/Flowise, Citizen Lab Webloc, lab-only AI-agent research: no constituency decision.
  • Reverted update: the SonicWall CVE-2026-19868 entry received an update record citing CERT-FR AVI-1507 on SNWLID-2026-0021 as a second SonicOS fix (7.3.4-7015); verifier iteration 1 read the bulletin through SonicWall's PSIRT API and found it is the informational eMudhra DPI-SSL advisory, already carried by the eMudhra entry, so the entry file was reverted to its base and removed from updated_entry_ids[].
  • Systemic: a short out-of-slice check moved datadog-security-labs' rotation cursor on 2026-11-12 (fixed in v4.25); entry-shape did not check sources an update adds (fixed in tools/check_run.py, with a test).
  • Registry: ambiguous_labels added for SRG on actor:silent-ransom-group and Cereal on malware:cereal; six product records from sync_products.py.
  • Source health: connectwise-trust-center and bfv-verfassungsschutz-de repaired and now relevant; safeonweb-be answers HTTP 500 site-wide and stays the one UNSOLVED record. The five essential sources with no contribution all week read cleanly in the re-sweeps.
  • Reader pool (context): every jina key exhausted.
  • ATT&CK pin: up to date (local v20.1 == upstream v20.1).
  • Verifier loop: seven iterations; truth and editorial findings fell 18, 9, 3, 4, 5, 0, 0; iteration 6 CLEAN on a 101-claim post-fix pass, confirmed by iteration 7's full cold pass over all 183 claims. Iteration 5 found a defect in this fire's Check Point section outside its sampled scope, so a post-fix pass's quarter sample can miss what the confirmation pass then has to catch. Three low-confidence advisories from iteration 7 are left for the next audit (Chrome Exposure citation scope, Dixell takeaway citation, the declined file-name point).
  • Budget: 16 spawns (6 truth passes, 3 re-sweeps, 7 verifier iterations), 8,667,965 sub-agent tokens (work/2026-11-15T1407Z-audit/budget.tsv).
  • Coverage gaps: sygnia (blocked on every transport), nict-cynex-lettice (JS shell, no reader credit), belden-psirt (no content on any transport), lab52 and anssi-cyber-gouv-actualites (no working transport this audit), the EFS press page and the SentinelLABS post (unreadable to TA).

2026-11-15T0356Z-intel · Sonnet 5.5 · window 28 h · 3 entries published

Verification & coverage notes

Window: gap 25.0 h (daily cadence, previous intel fire 2026-11-14T0255Z), window 28 h; a Saturday-to-Sunday window with the home-region feeds quiet (CERT.at newest 2026-10-21, BACS press 2026-11-05, NCSC Security Hub post 13178 of 2026-11-13). No outage backfill was triggered (the gap is the normal daily cadence); S3 swept the majors' listing pages for posts since 2026-11-13 anyway. KEV sweep: 0 additions in the window, no ransomware-flag flip on a covered CVE. Source slices were ranked on the rotation cursor alone; the previous-two-fires exclusion was not applied because it left the breaches slice empty.

Published: three new entries (PolinRider, ScreenConnect crypto-exchange bait sessions, The Gentlemen affiliate report) and one update record on the Cursor marketplace entry (Cursor's own report of 2026-11-13 gives the origin the entry said nobody had published). Deep dive: none, nothing cleared the bar.

  • Single-source: 2026-11-15/screenconnect-crypto-bait-guest-sessions-118-servers (Huntress alone, hypothesis stated as unobserved, ConnectWise bulletin unreadable); 2026-11-15/gentlemen-affiliate-vpn-no-mfa-da-gpo-locker-failed (Ransom-ISAC, reliability C, no victim named).
  • borderline-drop: jDownloads CVE-2026-117551 (Joomla), needs upload permission, no exploitation, Swiss footprint unknown.
  • borderline-drop: Fortinet FG-IR-26-187 revision of 2026-11-13 (Brotli decoder flaw), product-status list only, no fix or workaround, advisory not cited by the covered entry, so no patch decision changes.
  • borderline-drop: Qubole Wiz update of 2026-11-11; the "n0n" claim is relayed from a newsletter and unverified, the separate key-testing is already in the entry's sourcing note, no deployment of the bootable image is known.
  • borderline-drop: GeoVision camera software abuse (Huntress, 13 organisations), initial access unnamed, no CVE, no region; dropped on 2026-11-14 for the same reason.
  • borderline-drop: Unit 42 CL-STA-1238 (Cairn) and Elastic, Talos and Arctic Wolf agentic-framework internals (ARTEX, AntiFox), US-centred targets, class already covered by the ARTEX and CyberStrikeAI entries; dropped on 2026-11-13 and 2026-11-14.
  • borderline-drop: FBI FLASH and DOJ action on FaceStealer, Swiss nexus indirect; dropped on 2026-11-13 and 2026-11-14, no new fact.
  • borderline-drop: Project Zero "Full X" policy, effective 2027, no near-term decision.
  • borderline-drop: Elastic SANSAR (Ray job agent), CloudSEK Silent Lynx (Central Asia), Recorded Future survey (vendor survey), AMD SEV-SNP Mior and Skyfall (hypervisor-level attacker, CVSS 5.9).
  • borderline-drop: React Server Components CVE-2026-126187, application-code preconditions, no exploitation or proof of concept; dropped on 2026-11-12 and 2026-11-13.
  • out-of-window: Saxony emergency pager exposure (heise 2026-11-12, Spiegel 2026-11-11) and the Operation Endgame round (Europol 2026-11-12), window_hours=28; both were also dropped by the 2026-11-13 and 2026-11-14 fires, the pager item for lack of an access vector or actor.
  • borderline-drop: Quantum Systems QBase update-server compromise (sources 2026-11-13, no Swiss customer), Mittwald MCP-server intrusion (German host, post-mortem pending), Fireworks AI incident (folded into the Cursor update), Microsoft 365 outage of 2026-11-12 and 2026-11-13 (capacity upgrade, not an attack).
  • Backlog: the one open row (ChimeraZ claim against the Belgian rescue zone Hainaut-Est) struck on 2026-11-15: six French-language searches found no confirmation and the cited page does not mention the zone.
  • Coverage gaps: belden-psirt (403 on every transport); sygnia (403 on every transport); lab52 (connection resets, reader pool dead); anssi-cyber-gouv-actualites (Incapsula challenge); databreaches-net article pages (403, feed works); fbi-gov, quantum-systems.com, beobachter.ch and connectwise-trust-center HTML pages (403, other routes used); jina reader pool exhausted.
  • Watchlist: none configured, sweeps were no-ops.
  • Candidate sources considered and not added: norforge (no feed), fireworks-ai-security-bulletins (low volume, the Cursor update carries the bulletin), lemagit-cyberhebdo (weekly issue URLs change each week).
  • Mid-run stop-hook nudges to commit were declined; the run commits once, after the gate and the verifier loop.
  • Verification: three iterations, all NEEDS_FIXES (iteration 1: 11 truth, 1 editorial; iteration 2: 3 truth, 2 editorial; iteration 3: 1 truth, 0 editorial, a low-confidence wallet-attribution wording). The final residual of 1 was fixed after the pass and not re-verified; the loop stopped on the low-residual early exit. Iteration 1 caught two timestamps written ahead of the clock and a dropped hedge, both fixed.
  • The gate's dedup advisory on actor:thegentlemen is deliberate: the 2026-11-07 Kaspersky entry covers a different incident (a monitoring server's root password on ESXi and Proxmox), this one an SSL-VPN intrusion with a GPO locker failure.
  • Note for the next audit: the Cursor update was trimmed twice to stay within the 550-word analysis bound; the entry carries its original wording otherwise.