5 verified findings from 1 run · 4 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
01Kaspersky: BlackNevas used vCenter Host Profiles to enable SSH on ESXi, destroyed backups first, then wiped datastores. Kaspersky GERT describes a ten-day engagement in which an adversary, identified as BlackNevas from its ransom notes, used credentials taken from systems outside the virtualisation layer to reach both vCenter appliances, persisted on one with a remote-access agent and a tunnel installed as a VMware-lookalike systemd service, then used vCenter Host Profile remediation to switch SSH on across four ESXi hosts, destroyed the backup environment first and detonated a wiper-capable locker that took every virtual machine. Kaspersky states no initial access vector, victim sector or country, and no virtualisation-layer vulnerability was exploited. →
02GitLab critical patch: a merge request can reach a Git heap overflow on Gitaly; update to 19.5.4, 19.4.6 or 19.3.8. GitLab shipped an ad-hoc critical patch release on 2026-11-10 for CVE-2026-105437, a heap overflow in Git that, per GitLab, any user who can create a project and open a merge request reaches, and that can run commands as the git user with access to every repository on that Gitaly node. Git 2.56.1 and its backports fix the same flaw and five others, among them CVE-2026-111760, code execution when a malicious repository is cloned with --recurse-submodules; no source reports exploitation. →
03Red Hat Quay: anonymous signing-key takeover and cross-tenant token theft; update to your branch's fixed release. Red Hat's CNA records say an unauthenticated client of Red Hat Quay can obtain an anonymous registry token, present it to the service-key rotation endpoint and install its own signing key (CVE-2026-123710, CVSS 3.1 9.8), and that an account that can create a namespace containing an underscore can retrieve other tenants' robot tokens (CVE-2026-110697, 9.9). Red Hat fixed both in Quay 3.9.28, 3.10.28, 3.12.24, 3.14.10, 3.15.9, 3.16.7, 3.17.7 and 3.18.3, lists Quay 5 and the mirror registry for Red Hat OpenShift 2 as affected with no fix, and NCSC Switzerland lists exploitation as unknown. →
Kaspersky GERT describes an engagement in which an adversary destroyed an organisation's virtualisation estate over ten days: four ESXi hosts, every virtual machine on them, both vCenter Server Appliances and all connected backup infrastructure (Kaspersky GERT, 2026-11-13). No virtualisation-layer vulnerability was exploited: credentials obtained from non-virtualisation systems inside the perimeter reached the appliances, and every action against ESXi ran through legitimate VMware management interfaces (Kaspersky GERT, 2026-11-13). The activity is attributed to BlackNevas through the ransom notes (Kaspersky GERT, 2026-11-13).
The attackers logged in as root over SSH to the vCenter appliance that manages virtualisation disaster recovery, fetched and deployed MeshAgent with a shell script 43 seconds later, and about four hours later returned through MeshAgent to install the Ligolo-ng tunnel as a persistent systemd unit whose name, description and Restart=always imitate appliance housekeeping (Kaspersky GERT, 2026-11-13). They then used vCenter Host Profiles to change the TSM-SSH service state: eleven minutes after the first configuration read SSH was enabled on the target ESXi hosts, and direct SSH to the hosts then came from the vCenter appliance address (Kaspersky GERT, 2026-11-13). Payloads went to ESXi over SFTP, a self-test of the executable ran about three hours before the final detonation, and about the first 2 KB of the virtual disk files were overwritten (Kaspersky GERT, 2026-11-13). The backup environment fell first: the attacker extracted stored credentials after gaining administrative execution on the backup infrastructure, disabled antivirus there days earlier, then stopped backup services and encrypted the recovery infrastructure, all before ESXi detonation (Kaspersky GERT, 2026-11-13).
Triage: each Host Profile step is legitimate alone, so the distinguishing context is authorisation, timing, source account, change window and whether the desired state weakens security; a VMware-looking service is not malicious by default, since custom integrations exist, but one absent from the trusted list is a high-confidence anomaly, and chmod +x is ambiguous alone while execution of a newly staged executable from /tmp is itself a strong indicator (Kaspersky GERT, 2026-11-13).
recently investigated an engagement in which an adversary destroyed an organization’s virtualization estate over a 10-day period.
Subsequently, direct SSH access to the target ESXi hosts originated from the vCenter appliance IP, a source that would normally be considered highly trusted on the management network.
In this incident, backup infrastructure was not collateral damage.
ReliaQuest says Luna Moth, also called Silent Ransom Group, has targeted law firms since at least mid-2023 and in October and early November went back to firms it had extorted in May and August; those firms appear on the group's leak site, which ReliaQuest says suggests they likely did not pay, and the group exfiltrated a second round of data (ReliaQuest, 2026-11-12). The approach reverses the group's earlier flow: a blank-subject email from a free address carries a deliberately invalid PDF that crashes the viewer, and the attacker then calls the target, often on a personal number outside corporate call controls, posing as IT support responding to a "malicious email" (ReliaQuest, 2026-11-12). By late August the group paired video calls with a built-in remote-assistance tool, and by October the video call alone carried both the conversation and the screen control (ReliaQuest, 2026-11-12).
On the call the attacker has the employee open Microsoft 365 Copilot in the browser on a managed, compliant laptop, open the browser's developer tools to find the request Copilot sends to Microsoft Graph and copy the access token it carries, and paste the token into a public text-sharing site where the attacker retrieves it (ReliaQuest, 2026-11-12). Within a minute the attacker replays the token from its own server and downloads files en masse through the Graph API, thousands in about an hour (ReliaQuest, 2026-11-12). The stolen token passes Conditional Access because the device check runs only at sign-in, not when the token is later used, and, unlike device-code phishing, the token is issued to the victim's compliant device and then handed over (ReliaQuest, 2026-11-12). Nothing is installed, and the video-call applications are signed legitimate software (ReliaQuest, 2026-11-12).
This token-based exfiltration is almost certainly a new tactic for the group.
Attackers never have to authenticate, so no sign-in events are generated for their address.
In the observed intrusions, the first download from attacker infrastructure came about a minute after the token was copied.
Between 2026-09-30 and 2026-10-16 one server serving a CyberStrikeAI web interface used the credentials of eight AWS identities (five IAM users and three EC2 instance roles) across four accounts and generated 78 alerts, the first case Elastic has seen of this tooling turned on AWS (Elastic Security Labs, 2026-11-13). It started by validating identities: GetCallerIdentity on four IAM users in three accounts within 51 minutes, three of the calls in the same second (Elastic Security Labs, 2026-11-13). An instance role then read a secret, made one user-data request and fired 56 more in 141 milliseconds, spacing Elastic says fits tooling working through a prepared list of instance IDs (Elastic Security Labs, 2026-11-13). On 2026-10-02 another instance role read an encrypted Systems Manager parameter and issued a SendCommand 29 seconds later, which Elastic calls the highest-risk activity; the audit log does not show the command or its outcome (Elastic Security Labs, 2026-11-13). After 14 quiet days the server returned and revalidated credentials, including a user in a fourth account (Elastic Security Labs, 2026-11-13). Elastic does not know how the credentials were obtained and infers that the instance-role sessions, used from an outside address, were taken from the instances, for example through SSRF or host compromise (Elastic Security Labs, 2026-11-13). Team Cymru counted 21 unique addresses running CyberStrikeAI between 2026-01-20 and 2026-02-26 (Team Cymru, 2026-03-02).
Triage: each call is one a legitimate role makes every day, so the discriminators are the source address, the number of distinct identities, accounts or instances one address touches in minutes, and the decrypt-then-SendCommand pairing (Elastic Security Labs, 2026-11-13).
The main activity came from app-role, an EC2 instance role in Account A. It read a secret, made one user data request, then fired a burst of 56 user data requests in 141 milliseconds.
Only the shared source IP showed they belonged to one intrusion.
We cannot prove an LLM agent was making the decisions, and we do not know how the attacker obtained the credentials.
GitLab shipped an ad-hoc critical patch release on 2026-11-10, 19.5.4, 19.4.6 and 19.3.8 for CE and EE, for CVE-2026-105437, a vulnerability in Git that under certain conditions gives out-of-bounds heap writes that could run arbitrary commands in Gitaly's environment; it affects all versions from 16.2 before 19.3.8, 19.4 before 19.4.6 and 19.5 before 19.5.4, while GitLab.com and Dedicated already run a patched version (GitLab, 2026-11-10). GitLab's detection page says any user who can create a project and open a merge request reaches the flaw, which gives command execution as the git operating-system user and read and write access to every repository on that Gitaly node (GitLab, 2026-11-12).
Git 2.56.1 and the backports 2.55.1, 2.54.1, 2.53.1, 2.52.1, 2.51.3 and 2.50.2 of 2026-11-10 fix the same flaw and five others (GitHub Blog, 2026-11-10). The one that matters on workstations and CI runners is CVE-2026-111760: a long path that passes validation is written as .git and replaces a freshly cloned submodule's .git file with a symlink to a directory that holds attacker hooks or configuration, which GitHub says can execute attacker-controlled code during a recursive clone (oss-security, 2026-11-11; GitHub Blog, 2026-11-10). The maintainers believe it is only triggerable with --recurse-submodules and note that such a repository fails fsck, which does not protect downstream users, since an innocuous-looking .gitmodules pointing at it suffices (oss-security, 2026-11-11). GitHub Enterprise Server fixes are in 3.19.15, 3.20.11, 3.21.9 and 3.22.4, and GitHub is coordinating releases of GitHub Desktop and the Copilot app with patched Git (GitHub Blog, 2026-11-10); Git for Windows 2.56.1.windows.1 addresses the issue (Git for Windows, 2026-11-10). No source reports exploitation; GitLab documents a lab reproduction in which opening one merge request left a crash entry within a second and a half (GitLab, 2026-11-12).
In GitLab, under certain conditions, this vulnerability in git can result in out-of-bounds heap writes, which could be used to execute arbitrary commands within GitLab’s Gitaly environment.
Arbitrary command execution in the Gitaly environment, as the git operating system user, by any authenticated user who can create a project.
Red Hat's record for CVE-2026-123710 says an unauthenticated client of Red Hat Quay can obtain an anonymous registry token with an audience it chooses, present it to the service-key rotation endpoint and install its own approved signing key with no expiry (Red Hat, 2026-11-11). NCSC Switzerland summarises the impact as letting network attackers with anonymous or low-privileged access read, overwrite or poison private container images and steal registry credentials, compromising the software supply chain (NCSC Switzerland, 2026-11-13). The condition is that the /keys endpoint is reachable by untrusted clients (Red Hat, 2026-11-11); anonymous access is on by default, and turning it off does not protect, because any valid Quay account can still exploit the flaw (Red Hat, 2026-11-13).
CVE-2026-110697 lets an attacker who creates a namespace containing an underscore match and retrieve robot accounts and their decrypted tokens from other tenants' namespaces; Red Hat says it needs no elevated privileges or victim interaction and that, where Quay is OpenShift's integrated registry, stolen robot credentials can inject malicious images into CI/CD pipelines (Red Hat, 2026-11-11).
Red Hat's page for the second flaw lists the fixed builds 3.9.28, 3.10.28, 3.12.24, 3.14.10, 3.15.9, 3.16.7, 3.17.7 and 3.18.3 and says Quay.io is already fixed (Red Hat, 2026-11-11); its CSAF record for the first flaw links eight Quay errata, RHSA-2026:99019 and 99021 to 99027, as its fixes and lists Red Hat Quay 5 and the mirror registry for Red Hat OpenShift 2 as affected with no fix available (Red Hat, 2026-11-13). NCSC Switzerland lists the exploitation status as unknown (NCSC Switzerland, 2026-11-13).
An unauthenticated client can obtain an anonymous registry token with an audience it controls, present that token to the service key rotation endpoint, and install its own approved signing key with no expiry.
Update to the fixed Red Hat Quay release for your version: 3.9.28, 3.10.28, 3.12.24, 3.14.10, 3.15.9, 3.16.7, 3.17.7, or 3.18.3.
NCSC-NL reports that public proof-of-concept code for the unauthenticated SQL injection now exists, which it says raises the chance of exploitation in the short term, though it names no source and Ivanti says it has no evidence of exploitation; CERT-FR names Cloud Services Application 5.0.8.1 as the fixed build, so the update action now covers both products.
NCSC-NL revised its advisory on 2026-11-13 (revision 1.0.1) to say that public proof-of-concept code for CVE-2026-121653 can now be found on the internet and that this raises the chance of active exploitation in the short term; it rates probability medium and damage high, and names no source for the code (translated from Dutch) (NCSC-NL, 2026-11-13). CERT-FR's advisory of 2026-11-12 lists the Cloud Services Application as affected in version 5.0.8 before 5.0.8.1, so 5.0.8.1 is the fixed build (CERT-FR, 2026-11-12). Ivanti's blog of 2026-11-10 says it has no evidence of these vulnerabilities being exploited in the wild (Ivanti, 2026-11-10).
N-able confirmed via Huntress that Agent2Shell, the researcher's other chain set that gives an unauthenticated attacker SYSTEM commands on managed Windows endpoints, also needs the non-default Legacy API, and N-able has published indicators of compromise for customers that have not disabled it; N-able still has no fix, and its earlier statement that it saw no signs of compromise predates the indicators.
The researcher's 2026-11-12 disclosure also includes Agent2Shell, which The Hacker News describes as more than ten zero-day flaws used to run commands as SYSTEM on managed Windows agents, assuming the attacker knows the target appliance ID (The Hacker News, 2026-11-12). Huntress describes it as a set of more than ten issues in how the N-central server communicates with Windows Agents and Probes, two chains of which let an unauthenticated attacker who can reach the server run commands on managed Windows endpoints, including as SYSTEM (Huntress, 2026-11-13). Huntress relays that N-able confirmed Agent2Shell, like 4EverAble2Shell, requires the non-default Legacy API to be enabled, which settles the question Huntress and Tenable had left open on 2026-11-12, and that on 2026-11-13 at 13:40 UTC N-able said it had identified indicators of compromise for customers that have not disabled the Legacy API (Huntress, 2026-11-13); the statement that N-able had seen no signs of compromise dates from 2026-11-12 (Tenable, 2026-11-12). Huntress has not observed exploitation of these issues in its partners' environments, and its Agent2Shell indicators come from static analysis of the proof of concept, not from attacks seen (Huntress, 2026-11-13). Tenable adds that N-able has temporarily deactivated the Intune integration for on-premises customers running it (Tenable, 2026-11-12).
Microsoft's analysis describes RelayTrap, a rogue client built on the endpoint to deliver a payload to the technician who connects, and browser-cookie routines designed to push tools through the Shared Toolbox; ConnectWise says clients 26.6.5 and later block the transfer, and a second wave on 2026-11-13 hit 205 hosts at 41 organisations, one with a self-hosted RMM+, so every technician client now belongs on the action list.
Microsoft published on 2026-11-13 its analysis of SCKit, which includes RelayTrap, a malicious replacement for the ScreenConnect client on an affected customer endpoint that is designed to deliver a payload to a technician's workstation when they connect for support (Microsoft, 2026-11-13). A watcher on the endpoint enumerates the ScreenConnect client services and builds RelayTrap for each with the .NET C# compiler, and the toolkit's browser code searches Chrome, Edge, Opera and Opera GX profiles for cookies named .ASPXAUTH; a console-distribution component is designed to combine the cookie with the anti-forgery and access tokens, upload a payload, queue it through the Shared Toolbox and remove the uploaded item afterwards (Microsoft, 2026-11-13). Microsoft relays ConnectWise's statement that clients 26.6.5 and later block the transfer; in its own words the fix for CVE-2026-84869 was released on 2026-09-08 and a cloud-console update does not automatically update technician clients, and it advises live response or another out-of-band method instead of ScreenConnect on a suspected device, with connecting technician workstations in scope, invalidation of ScreenConnect web and identity-provider sessions, and no use of a cleanup script recovered from the kit (Microsoft, 2026-11-13). Blackpoint's own test found that a current Host client did not fetch or run the queued transfer, and it found no technical link between the extensions and RelayTrap or the browser routines (Blackpoint, 2026-11-13).
Huntress's update of 2026-11-13 reports a second spike at 16:12 UTC that hit 205 hosts at 41 organisations with no follow-on executions observed, one incident at an organisation running a self-hosted RMM+, a rogue client persisting as a service called Windows Update Agent, and ConnectWise's updated advisory urging partners to verify that every technician client runs 26.6.5 or later (Huntress, 2026-11-13). Blackpoint's update says the second wave appears to target on-premises instances where mitigations remain pending (Blackpoint, 2026-11-13).
Midnight Blue reported on 2026-11-11 six pre-authentication remote-code-execution flaws in all Danfoss AK-SM 800A firmware up to 5.3.2, with no CVE identifier, no fixed version and no known exploitation (BSI); CCB says updating, new passwords and Strict mode are insufficient and removing internet exposure comes first, and BSI will not rule out flaws in firmware 5.4.0 (heise).
Midnight Blue published on 2026-11-11 a report on six pre-authentication remote-code-execution flaws in the Danfoss AK-SM 800A, which it says it found within an hour of manual investigation and which affect all versions up to and including 5.3.2 (Midnight Blue, 2026-11-11). It says the exploited application runs as root, that the controller does not use the processor's secure-boot features, which makes malicious firmware that survives a factory reset or bricks a device feasible, and that the strict session-control mechanism also has at least one trivial pre-authentication memory-corruption flaw; it recommends against spending scarce effort on patching and in favour of reducing exposure (Midnight Blue, 2026-11-11). No CVE identifiers exist yet and technical details are withheld during coordinated disclosure with Danfoss (Midnight Blue, 2026-11-11).
BSI's notice of 2026-11-13, criticality 3 (Orange), lists no fixed version, says nothing is known about active exploitation and warns that the report may give attackers enough hints; its worst case is a remote attacker loading malicious firmware that leaves the devices unusable until they are replaced or restored by hand (translated from German) (BSI, 2026-11-13). CCB's advisory of 2026-11-12 relays the research, says no patches were available and that removing direct public internet exposure must not wait for them, and says Strict mode does not eliminate the risk because the research identifies a flaw in the session-control mechanism itself (CCB, 2026-11-12). heise reports that BSI told it that it does not want to rule out vulnerabilities in firmware 5.4.0 of 12 November either (heise, 2026-11-13). Danfoss's advisories page lists no advisory on the six as of 2026-11-14 (Danfoss, read 2026-11-14), and the readme for software release 5.4.0 says an update from 4.x leaves the security settings unchanged and names no vulnerability (Danfoss, read 2026-11-14). BACS's statement of 2026-11-10 that updates fix the known flaws (BACS, 2026-11-10) predates the report and does not address these six.
Find every Danfoss AK-SM and AK-SC, Copeland xweb and Dixell xweb refrigeration controller, and any other refrigeration controller with a remote management interface because not every second-wave site's vendor is confirmed, in facilities you own or that a contractor manages for you and take its management interface off the internet now, restoring access only through a secure VPN or protected network (filtering to the contractor's source addresses is an emergency stop-gap, and CERT-FR asks you to check your insurance conditions before cutting the contractor's data uplink); an end-of-life unit that gets no fixes needs replacing, because a factory reset leaves it vulnerable and probably compromised again.
On every such controller that was reachable from the internet, check the whole configuration, defrost ranges, alarm thresholds and notification recipients included, against the intended one, verify the actual temperatures and test that alarms reach staff; for a Danfoss unit that may have been taken over, restart it or otherwise factory-reset it, restore access only through a secure VPN or protected network and reconfigure it for service, then update its firmware, set the security level to Strict and replace the default and any weak passwords, which do not close the six unpatched AK-SM 800A flaws that Midnight Blue and CCB report, so the interface stays off the internet; change all credentials where compromise is suspected and check the devices attached to the controllers, as BSI advises; and on every Copeland or Dixell XWEB unit confirm the factory Admin password was changed and give each contractor a named account with an expiry date.
On every ScreenConnect instance your organisation or its IT providers run, disable the RMM+ extension immediately, verify it stays disabled and do not re-enable or reinstall it (on-premises instances by hand; ConnectWise removed it from cloud-hosted ones), review Run Command and Toolbox history and the instance audit logs for executions nobody started, including before 2026-11-11, and check managed endpoints for a second ScreenConnect client service; open no interactive session to a suspect host, use EDR live response or WinRM, because a session can trigger the toolkit to spread to the technician's workstation; update every technician host client to ScreenConnect 26.6.5 or later, because a cloud-console update does not update them.
Upgrade every self-managed GitLab to 19.5.4, 19.4.6 or 19.3.8 now (self-compiled installs rebuild the Git in Gitaly and the Git at git.bin_path, or move the distribution Git to 2.56.1 or later) and, where that waits, limit who can create projects and sign up, because a user who can open a merge request reaches the flaw and code execution gives access to every repository on the Gitaly node.
Update every Red Hat Quay registry to its branch's fixed release (3.9.28, 3.10.28, 3.12.24, 3.14.10, 3.15.9, 3.16.7, 3.17.7 or 3.18.3) and, until then, restrict the /keys endpoint at the proxy or load balancer to trusted internal components; on any instance whose /keys endpoint untrusted clients could reach, review audit logs for service_key_rotate and service_key_delete events with a null expiration_date and remove unrecognised approved service keys, because deleting a rogue key is not durable before the fix and resetting passwords or regenerating robot tokens does not revoke it.
Upgrade every Ivanti Endpoint Manager Core server to 2024 SU8 or later now and every Cloud Services Application 5.0.8 to 5.0.8.1, because NCSC-NL reports public proof-of-concept code for an unauthenticated SQL injection that reaches code execution on the server that distributes software to the managed endpoints.
2026-11-14T0255Z-intel· Sonnet 5.5 · window 26 h · 5 entries published
Verification & coverage notes
Window: standard 26 h (gap to the previous fire 24.0 h). KEV sweep: two additions in the window. CVE-2026-123039 (Check Point) is covered by an entry that already says exploited, so the listing is bookkeeping and became an internal record that adds the cisa-kev status; CVE-2026-114713 (UnrealIRCd) is a borderline-drop below. No ransomware-flag flips on covered entries. No closed-source drops present (S5 not spawned). Five new entries and five changelog records (four updates, one internal improvement); no deep dive and no critical. The will-publish set exceeded eight items, so the Phase 4 deep read was delegated to five scoped cti-research workers (DR-A to DR-E, saved bodies under work/<run-id>/dr/); three of them did not report call telemetry (null above). Phase 0 deviations: the entity registry (738 KB) was searched with grep rather than read whole, and of the 100 prior-coverage records the 48 newest were read in full and the older 52 as title, headline and key projections (those were last active before 2026-11-11).
Backlog: the ChimeraZ / Hainaut-Est (ZOHE) row stays held; S2 and S4 searched in French, Dutch and German and found no confirmation, expiry 2026-11-17 unchanged.
Inclusions with reduced confidence or carve-outs:
Luna Moth token handoff (notable): single vendor source (ReliaQuest, which sells the platform it recommends), no independent confirmation; credibility B3, verification: single-source; admitted through the rotational-lookback rule (d) and PD-11(d).
CyberStrikeAI against AWS (notable): Elastic is the only source for the intrusion, Team Cymru documents only the tool; Elastic does not know how credentials were obtained.
Kaspersky BlackNevas vCenter case (notable): single vendor source; victim, sector, country and vector unstated; kept distinct from the 2026-11-07 Kaspersky case in references[].
Red Hat Quay (notable): vendor CNA records plus NCSC.ch restatement; a third-party repository claims a lab proof of concept for CVE-2026-110697 and is not linked; exploitation: unknown.
GitLab critical patch and Git 2.56.1 (notable): admitted under the out-of-band limb of PD-11(b) (GitLab's ad-hoc critical patch release); CVSS 7.5 in GitLab's CNA record against a Critical label on its patch page, printed as stated.
Ivanti EPM update: NCSC-NL's statement of public PoC code is the only source and names none; the only repository found is a third party's lab model; priority kept at notable.
borderline-drop lines:
borderline-drop: CVE-2026-114713 UnrealIRCd 6.0.0 to 6.2.7 (KEV 2026-11-13, vendor says exploited), no constituency nexus (IRC networks with two or more linked servers); doubt about relevance resolves to drop.
borderline-drop: PostgreSQL 18.7/17.12/16.16/15.20/14.25 (52 CVEs, 32 at 8.8; PostgreSQL 14 final release), scheduled minor release inside the regular cycle, 43 of 52 need a low-privileged SQL role, no exploitation or public PoC.
borderline-drop: Node.js November 2026 security releases (16 CVEs, 7 high), regular release; each high needs a specific handler pattern; no exploitation.
borderline-drop: AXIS Camera Station CVE-2026-13312 (9.9), dropped again: no mechanism published, no exploitation, fix for Camera Station 5 planned for November.
borderline-drop: Fortinet FG-IR-26-268 revision (IPS virtual patch), no exploitation and no material delta for a flaw the entry already tells readers to upgrade for.
borderline-drop: Microsoft Patch Tuesday "Exploitation More Likely" network RCEs (Schannel CVE-2026-119952, NAT CVE-2026-113862, 8.1, high complexity), would not change the decision (install the November cumulative update).
borderline-drop: Operation Endgame round against Gootloader, CastleRAT and StealC (Europol 2026-11-12), no Swiss agency or victim named; narrow defender consequence.
borderline-drop: Microsoft 365 / Exchange Online global outage of 2026-11-12/13, not an attack; root-cause report pending.
borderline-drop: Saxony emergency-service pager exposure, out-of-window: heise 2026-11-12T15:19Z precedes the window start 2026-11-13T00:57Z, already dropped by the previous fire's verification; Swiss pager status unverified.
borderline-drop: Quantum Systems QBase update-server compromise, no Swiss customer shown, access route and malware function unpublished, no actor named, core facts rest on unnamed sources; re-check on a Swiss nexus or a technical disclosure.
borderline-drop: Elastic and Talos agentic-framework internals (ARTEX, CyberStrikeAI, Cairn, PentAGI, HexStrike), no campaign news; ARTEX behavior already covered; the concrete hunt is in the CyberStrikeAI entry.
borderline-drop: Huntress fake crypto-exchange sessions on 118 self-hosted ScreenConnect servers, no malicious activity observed; the bait hypothesis for CVE-2026-84869 is unobserved.
borderline-drop: FaceStealer (FBI FLASH / DOJ), DPRK crypto-theft scheme against developers; Swiss public-sector nexus only indirect.
borderline-drop: SOCRadar XaocMail, Phantom53 and Havoc, single C-reliability vendor source, no victims, no corroboration.
borderline-drop: GeoVision camera software abuse (Huntress, 13 organisations), initial exploit unnamed, no CVE, cryptominer impact.
borderline-drop: Project Zero "Full 90" disclosure policy, vendor policy effective 2027, no near-term patch or hunt decision.
borderline-drop: Sysdig on CVE-2026-119618; the covered entry already states the public crash test without code execution.
borderline-drop: Cisco BroadWorks CVE-2026-20320 and Openwave CVE-2026-111793 (KEV 2026-11-10), drop stands; Cisco's v1.2 of 2026-11-13 says no direct evidence of exploitation.
borderline-drop: Mitel MX-ONE MISA-2026-0011, Zoom ZSB-26023/-24, Adobe AEM APSB26-189, OneNote CVE-2026-126526, .NET November servicing, Chrome 157/158 rows, the Siemens and CISA ICS November set, Synology, Zammad, n8n, NATS, OpenCTI, Linksys, Edito CMS, Palo Alto Low and Medium batch, S1 dispositions stand: authenticated, user-interaction, LAN-only, already dropped by earlier fires, or below the bar.
borderline-drop: Poland ePUAP, Maileva 22M claim, Mittwald MCP intrusion, Softway Medical, CHU de Reims, French communes, Trezor, Widerøe, TSMC, Curious Worm, ICO Somerset, S4 dispositions stand: out of nexus, no vector or actor, single-outlet claims, or not a breach.
Notes for the next audit: the GitLab work-item pages that DR-D fetched returned site boilerplate containing instructions addressed to AI agents; the worker ignored them and nothing from those pages was used or cited. The Danfoss entry's main analysis was rewritten from 702 to under 550 words (the recap of earlier waves lives in its update sections). Danfoss, ScreenConnect and N-able entries were already at or near the 550-word bound, so each delta replaced earlier sentences instead of adding to them. postgresql-security was proposed by S1 as a candidate source and not added. Midnight Blue's own post is not cited as primary on the Danfoss entry because it is a discoverer's report relayed by two authorities.
Verification: three iterations of the single cold-reader definition, none CLEAN. Iteration 1 returned 13 truth, 3 editorial and 4 advisory findings (attribution drift, a dropped hedge, citation dates, one mis-parented process claim), iteration 2 returned 3 truth and 2 advisory, iteration 3 returned 1 truth and 1 advisory with no F1 or F4; every claim of the ledger was walked on each pass and the full gate was green before the first spawn. The loop ended under the low-residual early exit (truth plus editorial at most 2, no F1 or F4): the one remaining finding, a GitLab clause that needed the maintainers' announcement cited next to GitHub's, was fixed after iteration 3 and the entry was not re-verified, so verification_residual_count is 1. The Ivanti priority finding (notable against high) was declined with a rebuttal recorded above and accepted by the iteration-2 verifier. Two advisory findings stand unchanged: the N-able and ScreenConnect titles and summaries have accreted across records, and the Danfoss second action is long.
Self-evolution: the end-of-run health sweep flagged cisa-news (a fetch_method: webfetch record) as needing a bridge because its content check was read through the reader while reachability was ua-blocked, which the v4.18 webfetch-only rewrite does not cover. tools/source_health.py gained _webfetch_reach_wall, tools/test_source_health.py a test (15 of 15 pass), and a second sweep ended with no unsolved record. No prompt, agent definition or site file changed, so no prompt version bump applies.
Coverage gaps: connectwise-trust-center (403 on every transport; the CNA disclosure repository was read instead); sygnia (403 on every transport again); lab52 (every direct transport reset the connection); belden-psirt (403); schneier atom feed (403, recovered through extract of the home page); ox-security article pages (403, feed readable); uptime.n-able.com (script shell: N-able statements after 2026-11-10 are as relayed); quantum-systems.com statement (403); ico-uk was absent from S2's slice and was attempted by S4 (recipe ico-uk enforcement 15 works again).