Window: approximately 16 November 09:00 SGT to 17 November 09:00 SGT (24 hours). New and overnight developments, deduplicated against the 16 November edition, with gaps stated plainly.
Barracuda confirmed on 16 Nov that an attacker used a cloud access key it had issued to a third-party AI company and never retired — the key sat in a configuration file in the vendor’s GitHub repository. About 10 cloud accounts were entered over 44 hours on 1–2 November; the intrusion matches the ARTEX pattern.
Infoblox confirmed on 16 Nov that attackers used a third-party data-migration vendor’s trusted access to exfiltrate data from its Microsoft SharePoint — customer NIOS configuration data and employee records including Social Security numbers and direct-deposit details.
N-able released Hotfix 3 on 16 November for the 4EverAble2Shell / Agent2Shell chains; the Legacy API is now permanently blocked. Hosted upgrades are complete and on-premises general availability is open. Huntback counted 21 N-central servers mining XMRig, 19 of them already on Hotfix 2.
Catalogue still 1,781 entries (2026.11.13). No additions on 16 or 17 November. The two 13 Nov additions (CVE-2026-123039 Check Point, CVE-2026-114713 UnrealIRCd) fell due on 16 November; neither is carried as a new row — a due date alone is not a new development under the standing rule.
Two confirmed supply-chain breaches in one day, both through trusted third parties — and Google says agentic attacks have crossed a threshold. Barracuda confirmed on 16 November that an attacker used a cloud access key it had issued to a third-party AI company and never retired — the key sat in a configuration file in the vendor’s GitHub repository — to enter about 10 cloud accounts over 44 hours on 1–2 November, in an intrusion matching the ARTEX pattern. Infoblox confirmed the same day that attackers used a data-migration vendor’s trusted access to exfiltrate customer NIOS configuration data and employee records, including Social Security numbers, from its SharePoint. Google’s AI Threat Tracker, published overnight, says adversarial AI has reached a “critical inflection point”: agentic frameworks are now used in live intrusions, and Mandiant responded to LLMJacking incidents in 2026 affecting financial services firms, with single-day fraudulent charges over $100,000 and worst cases reaching $1 million. The Barracuda failure was a key-lifecycle failure, not an AI-framework failure: a key issued for a retired project stayed live. The Infoblox “products not involved” line does not cover the data already taken: customer network-identity configuration in attacker hands is follow-on attack material.
Patching moves: N-able Hotfix 3 has shipped, Apache Axis exploitation is confirmed against two products, and ScreenConnect was exploited against MSP technicians. N-able released Hotfix 3 on 16 November for the 4EverAble2Shell / Agent2Shell chains and has permanently blocked the Legacy API; hosted upgrades are complete and on-premises general availability is open. Rapid7 confirms Apache Axis CVE-2026-90436 (9.8, unauthenticated remote code execution) was exploited against Atlassian Confluence Data Center on 13 November as well as against N-central, and Huntback counted 21 N-central servers mining XMRig, 19 of them already on Hotfix 2. Huntress confirms ScreenConnect CVE-2026-84869 (9.9) was exploited against MSP technicians: a guest session compromised a technician workstation in about nine seconds on 12 November, and more than 688,000 guest “run file” requests were sprayed across 42 MSP servers between 20 October and 16 November. The ScreenConnect fix sits on the technician client, not the server — patching the server alone leaves the hole open — and Hotfix 2 was not sufficient protection for N-central, so HF3 should not wait for a maintenance window.
Korea: the FSI special audit has begun, KOMSCO confirms 3,083 customers affected, and Coinone took about 70 hours to disclose — while N0n’s leak-site claims now name Visa, Mastercard and JPMorgan Chase. The Financial Services Commission opened its special audit of the breached Financial Security Institute on 16 November; the FSI’s evaluation system is still shut down with no recovery date. KOMSCO confirmed 3,083 customers affected in the chak gift-certificate platform breach and its president apologised; the attack ran through an API-authentication flaw. Coinone took about 70 hours 40 minutes from being told of the MetaM breach to its public notice — 80 minutes inside the 72-hour legal deadline — and the victim count is still undetermined. N0n’s leak-site listings now claim cardmember programme data from Visa, Mastercard and JPMorgan Chase, plus client records from D.R. Johnson Financial and Tailored Benefits — all unconfirmed, with no data samples and no company confirmations, and Bitdefender has shown N0n inflates its victim count. No new CISA KEV additions were made on 16 or 17 November.
Confirmed, reported and unconfirmed items are labelled as such. No MITRE mapping is shown where the source published none.
Security vendor; cloud accounts, source-code repositories and Kubernetes clusters
Networking and security vendor widely deployed in BFSI networks; customer NIOS configuration data and employee records
Irish stockbroker and wealth manager (AIB-owned); client email addresses
Managed service providers and their clients, including financial firms; technician workstations
Atlassian Confluence Data Center; N-able N-central servers
RMM platform used by MSPs serving small banks and financial firms
Financial-sector security assessment body, South Korea; assessment staff at banks, insurers, securities firms
Regional gift-certificate payment platform, South Korea (84 local governments, about 4.3 million members)
Crypto exchange; customer-support contractor MetaM, South Korea
Global card networks, the largest US bank, and US and Canadian financial advisory and benefits firms
Indonesian sharia bank; fund-transfer information and credit reports claimed
Brazilian credit lender and payments firm; payment-card photographs and identity documents claimed
US financial advisory and financial-services firms
No new CISA KEV additions were made on 16 or 17 November. The catalogue stands at 1,781 entries (version 2026.11.13). The two 13 November additions (CVE-2026-123039 Check Point, CVE-2026-114713 UnrealIRCd) fell due on 16 November, and neither is carried as a row: a due date alone is not a new development under the standing rule. SecurityWeek reported on 16 November that CVE-2026-114713 is being exploited in attacks, but no new attribution, IOC or PoC was published. The Check Point November Jumbo Hotfix federal due date passed on 16 November — apply it if not yet done (see Indicators & defensive actions).
| CVE ID | CVSS score | Affected product | Exploitation status (PoC / in the wild) | Relevance to finance |
|---|---|---|---|---|
| CVE-2026-84869 — NEW, exploitation confirmed | 9.9 | ConnectWise ScreenConnect, technician host client below 26.6.5 | Exploitation confirmed (Huntress, 12 Nov incident) | A guest in a ScreenConnect session triggers file transfers the technician’s host client runs without authorisation. Huntress confirms a full technician-workstation compromise in about nine seconds on 12 Nov and more than 688,000 guest “run file” requests across 42 MSP servers from 20 Oct to 16 Nov. In CISA KEV since 11 September. Fix: upgrade technician clients to 26.6.5+ — the server patch alone does not fix it — or remove the TransferFiles permission. A technician compromise can cascade to every client the MSP serves, including financial firms. |
| CVE-2026-116606 / CVE-2026-111560 — NEW alert (CSA AL-2026-162, 16 Nov) | 9.8 / 7.2 (CSA v3.1) vs 9.3 / 8.6 (Barracuda / watchTowr v4) | Barracuda Email Security Gateway, Load Balancer ADC, Message Archiver, WAF Control Center, WAF, Web Security Gateway | In the Wild (KEV since 10 Nov); public PoC | Actively exploited, chained: authentication bypass (CVE-2026-116606) with OS command injection (CVE-2026-111560) for unauthenticated root on an exposed Email Security Gateway admin interface. CSA Singapore issued alert AL-2026-162 on 16 November. The CVEs were covered in the 14 November edition; the CSA alert and the confirmed active exploitation are the new items. Do not merge the two scoring versions. A proof of concept was published by watchTowr Labs about 3 November, six days before Barracuda confirmed exploitation on 9 November against “a small number of organizations” — the repository URL was not verified first-hand and is not linked. Fix: ESG 9.4.0.030; CSA says update to the latest versions immediately. Email gateways sit in front of bank mail flow; unauthenticated root on one is a full-compromise path. |
| CVE-2026-99288 (“NoisyTCP”) — NEW | 7.5 (Red Hat draft) / 8.1 (Oracle) / Medium (Canonical); NVD unscored | Linux kernel TCP, 4.4+ (race condition in the lockless TCP established-socket lookup, latent since October 2015) | No public exploit — remote kernel panic demonstrated | Unauthenticated remote kernel panic (denial of service) via crafted TCP packets to any listening port, including Nginx, Apache and SSH defaults. No remote code execution was demonstrated. Full write-up published by researcher Hyunwoo Kim on oss-security on 15 November 16:00 UTC (16 Nov 00:00 SGT). Fixed kernels 6.12.117, 6.18.63, 7.3.4 (plus a second patch for arm64 / POWER / RISC-V); Debian, Oracle UEK and CIQ Rocky have shipped fixes, while RHEL 8/9/10 and Ubuntu had not as of 16 November. Not in CISA KEV. Linux servers and cloud infrastructure underpin bank systems; a remote unauthenticated panic against internet-facing hosts is a live availability risk. Do not merge the conflicting scores. |
| CVE-2026-129349 — NEW, catch-up (missed by the 16 Nov edition) | 9.1 | Apache Shiro through 3.0.0 (DefaultLdapRealm / JndiLdapRealm) | No confirmed exploitation; no verified public PoC | LDAP authentication bypass: with authenticationCachingEnabled=true (not the default), a user who logged in once can then log in with any password. Disclosed 14–15 November in a batch of 17 Apache flaws; fixed in Shiro 3.0.1. Aggregators list a GitHub repository (googlesky/CVE-2026-129349) as a public PoC, but it was not verified as genuine and is not linked. Shiro is a Java authentication framework used in enterprise applications and APIs. The non-default precondition limits exposure, but any deployment with LDAP caching enabled should patch. |
| CVE-2026-126259 — NEW, scope expansion 16 Nov | 9.0 | FortiOS 8.0.0, 7.6.0–7.6.7, 7.4.0–7.4.12, 7.2.0–7.2.13, and 7.0 all versions (added 16 Nov) | Not known exploited | Missing authentication (CWE-306): where HA FGSP with standalone-config-sync or autoscale is in use, an unauthenticated attacker can execute commands as super admin via crafted HA packets. Fortinet PSIRT FG-IR-26-273, first published 10–11 November; the 16 November update added FortiOS 7.0 all versions to the affected list. Fixed in 8.0.1 / 7.6.8 / 7.4.13 / 7.2.14; virtual patch FG-VD-61445.0day is available. FortiGate is edge infrastructure for banks; an unauthenticated super-admin path on HA pairs is a patching priority where the configuration applies. |
| 4EverAble2Shell / Agent2Shell — CVE-2026-90436 + VULN-2026-0009 — STATUS, HF3 shipped 16 Nov | 9.8 (CVE-2026-90436) | N-able N-central (on-premises servers with Legacy API enabled); Atlassian Confluence Data Center 1.x–9.x with Remote API enabled | Exploitation confirmed (Rapid7, 13 Nov) | Hotfix 3 released 16 November — phased rollout for hosted environments began that day, on-premises general availability is open, and the Legacy API is now permanently blocked. Until HF3 is applied, Hotfix 2 plus Legacy API disabled plus Agent / Probe upgrades remain the guidance. Rapid7 confirms CVE-2026-90436 exploitation against Confluence Data Center on 13 November and against N-central on 13 November; a PoC targeting N-central was published 12 November. Huntback counted 21 production N-central servers mining XMRig, 19 of them on the Hotfix 2 build. Apache Axis 1.x is end-of-life with no fixed release planned; Confluence is fixed in 9.2.28 and 10.x is unaffected. PoC repo: MSPNightmare/CVEs — N-Central N-Able (GitHub) — genuine researcher disclosure (Sean Bales), with Previdian’s caveat that the public PoC is unverified and may contain malware. RMM used by MSPs serving small banks; upgrade to HF3 now and hunt the IOCs in Indicators. |
| CVE-2026-121653 — STATUS, PoC status contested | 9.8 | Ivanti Endpoint Manager Core, 2024 SU7 and earlier | Public exploit reported — contested; in-the-wild exploitation not confirmed | No new development in-window. Unauthenticated SQL injection leading to remote code execution on the endpoint-management server. Fix: EPM 2024 SU8; no workaround published. Ivanti notes it is not externally exploitable in dual-NIC deployments and says it is not aware of any customers being exploited. The public-exploit picture is now contradictory: SANS (12 Nov) said no public working exploit, while a Week-46 roundup lists it as a public exploit with no verifiable repository — no genuine PoC repository was found, so no repo is linked and the PoC should not be presented as confirmed. VulnCheck added the CVE to its KEV feed on 15 Nov — that is VulnCheck’s feed, not CISA KEV; do not present it as a CISA listing. Not in CISA KEV. Sibling critical in the same 10 Nov batch: CVE-2026-121683 (9.4, missing authentication in Cloud Services Appliance 5.0.8 only; fix 5.0.8.1). EPM Core is the management server for bank endpoints; unauthenticated RCE on it is a full-compromise path. |
| CVE-2026-76466 — NEW, catch-up | 9.8 | Cisco Expressway, clustered Expressway-C / Expressway-E, X15.5.3 and earlier | Public PoC — exploitation not confirmed | Catch-up — Cisco advisory 21 Oct; PoC repository created 5 Nov; press coverage 12 Nov; never covered in a prior edition. Hard-coded credentials (CWE-798): a TLS 1.3 connection to TCP/4372, an Erlang distribution handshake as wombat@127.0.0.1 with the hard-coded cookie wombat, then rpc:call(os, cmd) as root. Fix: X15.5.20; X14 and earlier must migrate; no workaround. Validated only on Expressway-C X15.3.2, a two-node cluster in TLS Permissive mode (the factory default); TLS Enforcing mode is untested. Not in CISA KEV; Cisco PSIRT is aware of no malicious use. PoC: GitHub dinosn/cve-2026-76466 (Nicolas Krassas, created 5 Nov; exploit, fleet detector and docs — verified genuine). Do not link the Feedly-listed copy emirwtz/CVE-2026-76466 — not verified. Expressway is video-conferencing edge in banks; restrict TCP/4371–4372 to cluster peers until patched. |
| CVE-2026-126173 — NEW | 9.8 | AutomatorWP Pro Webhooks add-on ≤ 1.4.4 | Public PoC — in-the-wild exploitation not confirmed | Missing authorisation in the webhook handler lets an unauthenticated remote attacker reset any user’s password, including administrators. Published 14 Nov (Wordfence CNA); public PoC published the same day. PoC: GitHub im0xfay/CVE-2026-126173 (created 14 Nov; contains poc.py, a Docker lab and analysis — verified genuine, but it is a third-party reproducer, not the original discoverer’s repo; the README is in Indonesian and it includes mass-scan and mass-exploit modes plus a claimed live run against two FOFA targets on 14 Nov — the mass-exploit tooling is the risk escalator). No fixed version was listed at publication. PatchCVE’s claim of “public exploitation activity reported” is an unverified aggregator assertion — treat as unconfirmed. Not in CISA KEV. Bank marketing and content WordPress sites running this add-on face admin takeover, defacement and customer phishing. |
| CVE-2026-19907 — NEW | 9.8 | Shipo WordPress plugin 1.6–1.8.0 | No PoC reported | Unauthenticated admin authentication bypass: authenticate_rest_user() trusts a username taken from the LOGGED_IN_COOKIE without validating its signature, so knowing a username (for example admin) is enough. Disclosed 14 Nov (Wordfence CNA). No PoC reported; not in CISA KEV. The WordPress directory offers 1.8.4, but its changelog does not name this CVE — the fixed version is not independently confirmed. Same WordPress-site caveat as the AutomatorWP flaw, but no PoC and a narrower plugin. |
| CVE-2026-124969 — NEW, brief note | 7.1 | Optimole WordPress plugin ≤ 4.2.19 | No PoC reported | Unauthenticated arbitrary file deletion (via comment_content / comment_agent during attachment rename) that can lead to remote code execution by deleting wp-config.php; it needs PEAR Archive_Tar present and a subsequent admin rename action. Published 14 Nov. No PoC; not in CISA KEV. |
| CVE-2026-20320 — STATUS, vendor walk-back | — not stated in sources read | Cisco BroadWorks | In the Wild (KEV) — vendor evidence softened | Cisco advisory v1.2, dated 13 Nov, corrected the record: Cisco states it does not have direct evidence of active exploitation, walking back the 12 Nov update that reflected known active exploitation. The CISA KEV listing (added 10 Nov) stands. Treat the exploitation picture as contested between the KEV listing and the vendor’s own evidence position. |
| CVE-2026-109618 / 109619 (PHP2Shell) — STATUS, Siemens downstream notices | 9.2 | PHP; PAN-OS and other network/security products; Check Point Security Gateways | In the Wild | New in-window development: Siemens confirmed PHP2Shell affects RUGGEDCOM APE1808 running PAN-OS and several building-automation and energy products (SSA-566512; CISA ICSA-26-316-05) — industrial, not BFSI, so one line only. The Lazarus targeting discussion (Cyperma via Yonhap, 15 Nov 08:43 SGT) was covered in the 15 November edition; the attribution is Cyperma’s, reported not independently confirmed. Check Point’s detail adds that the TSGA/TSProxy implants are linked to the PhiliKit toolset (which ESET ties to UNC5221), but Check Point says it “cannot assess the operation with high confidence”. GreyNoise telemetry reported in the 13 Nov afternoon deep-research check shows PHP2Shell exploitation continued through 12 Nov. The Check Point campaign detail is unchanged: attackers chained PHP2Shell in the Mobile Access Portal with CVE-2026-123039 to gain root on unpatched gateways, planted the TSGA backdoor, pivoted to a VMware ESXi host and ran a wiper — “a couple of dozen” systems in one region. Both fixes are in the November Jumbo Hotfix. The PHP2Shell fix had been available since 23 Oct — the one-week patch lag is the lesson. Keep management interfaces off the internet; apply the November Jumbo Hotfix. IOCs: 130.94.19[.]204, 130.94.19[.]168; upload_progress_pwn string in PHP session files; process masquerading as CPconfigdaemon; DREAM_ANNEXATION.txt notes. |
| CVE-2026-121529 (“Fortinet2Shell”) — STATUS, PoC count up | Critical (no CVSS figure in sources read) | FortiClientMac 7.4.0–7.4.7; FortiClientWindows 7.4.0–7.4.7 and 7.2.0–7.2.14 | Public exploit — in-the-wild exploitation not reported | New in-window detail: the public PoC count is now five GitHub repositories per cve.now (the 13 November edition listed two). Still no confirmed in-the-wild exploitation; not in CISA KEV. Origin-validation flaw, unauthenticated remote code execution on FortiClient endpoints. PSIRT FG-IR-26-194. FortiClient is the endpoint agent on bank staff machines; an unauthenticated RCE with a public exploit is a patching priority. |
| CVE-2026-6460 — NEW, catch-up | 8.8 | PostgreSQL 14–18 (flaw dates to 2004) | Public partial PoC — in-the-wild exploitation not reported | Catch-up — patch published 12 Nov, researcher write-up 13 Nov; missed by the 14 November edition. A type-confusion flaw in PostgreSQL core (composite types plus ALTER TYPE between PREPARE and EXECUTE) lets an authenticated database user with object-creation privileges execute arbitrary code as the OS user running the database server. CVSS 8.8. It was one of 52 flaws fixed in the 12 Nov release (32 scored 8.8); the researcher traces the primitive to May 2004 — effectively every PostgreSQL since then. Researcher Mehmet Ince published the crash PoC and address-leak steps on 13 Nov and deliberately withheld the final RCE step, so there is a public partial PoC, not a copy-paste weapon. Full protection also requires rebuilding C-language extensions. PostgreSQL 14 is now end-of-life (14.25 is its last release). Sibling flaw CVE-2026-116322 lets an unauthenticated client trigger a restart of PostgreSQL 18 servers (denial of service). Not in CISA KEV. PostgreSQL underpins many banking and fintech stacks; authenticated RCE as the database OS user is a high-priority patch. |
| CVE-2026-119618 — STATUS, PoC verified as crash-only | 9.8 (Red Hat) / no CVSS yet (NVD) | Apache HTTP Server 2.4.55–2.4.70; standalone mod_h2 2.0.8–2.0.42 | Public PoC — in-the-wild exploitation not confirmed | No new development in-window. The third-party research PoC (GitHub googlesky/CVE-2026-119618, created 12 Nov) was verified directly on GitHub in the 14 November edition. It reproduces heap corruption and ASan crashes on 2.4.70 and clean behaviour on 2.4.71, with a containerised lab and ASan/TSan evidence — but it does not demonstrate code execution. Conflict — flag: Red Hat scores the CVE 9.8 and says public research demonstrates a reliable exploit against a default-configured server; Sysdig (13 Nov) and Abijita (12 Nov) both stress there is no public RCE PoC and no confirmed in-the-wild exploitation as of 12–13 Nov, and the CVE is not in CISA KEV. Treat exploitation status as “public PoC now exists; it demonstrates heap corruption, not code execution; in-the-wild exploitation not confirmed”. Nine GitHub repos now carry the CVE name, including two new labs on 13 Nov that were not verified — do not link them. Race condition in mod_http2 assign_headers(). Fixed in httpd 2.4.71; disabling HTTP/2 is a mitigation. |
| CVE-2026-124194 — STATUS, Patch Tuesday follow-up | 6.5 | Microsoft Outlook (NTLMv2 hash disclosure) | In the Wild (KEV) | No new development in-window. November Patch Tuesday is the largest on record by Qualys’s count — 1,252 CVEs this month (Qualys, 13 Nov). Conflict — flag: Microsoft’s own count differs (1,362 CVEs across 27 updates per other trackers) and Tenable counted 1,093 — a counting-method difference between trackers, not a new item. About 35% of November patches carried a Low reliability rating in Qualys’s assessment as of 11 Nov. Outlook CVE-2026-124194 remains the exploited item in the batch. Microsoft confirmed the Windows 11 update KB5132061 fails to install on some devices that took the October non-security preview KB5127936, and KB5131085 (Windows 10 / Server 2019) is reported to break Exchange — both affect deployment planning. |
| CVE-2026-127063 — carried | 9.3 | DataComm RMM+ extension for ConnectWise ScreenConnect | Not known exploited — no PoC reported | No change in-window. “The third-party RMM+ extension for ConnectWise ScreenConnect uses a static API secret by default.” Caveat: the CVE wording (“static API secret”) differs from Huntress’s account of a blank access key accepted when none was configured — treat as the same extension weakness pending vendor confirmation. This is the extension weakness behind the SCKit worm. Check RMM+ versions 1.0.31/1.0.32 and remove or re-key the extension. |
| Able2Shell — seven CVEs — carried | 9.3 (highest) | N-able N-central | PoC public — exploitation of new chains claimed, contested | No change in-window except the 4EverAble2Shell row above. Original Able2Shell (9 Nov): researcher Sean Bales published PoC code for 33 alleged N-central flaws. Hotfix N-central 2026.4 HF2 (build 2026.4.1.26), released 10 Nov, fixes seven CVEs: CVE-2026-125883 (XXE, 9.3), CVE-2026-125878 (OGNL injection), CVE-2026-125879, CVE-2026-125880, CVE-2026-125881 (SSRF), CVE-2026-125882 (RCE via path traversal), CVE-2026-91000. All hosted servers upgraded by 11 Nov 01:04 UTC. The Able2Shell PoC repository (Oracle-Security/CVEs, path “0-Days/N-Central N-Able”, 26 Python PoC scripts) is a genuine researcher disclosure — linkable as the PoC source, with the caveat that N-able has not validated all 33 findings. |
| CVE-2026-123864 — carried | High (no CVSS published) | Node.js 22.x / 24.x / 26.x (HTTPS servers) | Not known exploited — no PoC reported | No change in-window. HTTPS heap-memory disclosure: a zero-length TLS write re-enters the HTTP parser, and the server copies other connections’ heap memory — including TLS private keys and other users’ requests — into a request body. Fixed in 26.13.1 / 24.22.3 / 22.23.5. Node.js published no CVSS scores and lists no workarounds — upgrade only. |
| Node.js batch — six further High — carried | High (no CVSS published) | Node.js 22.x / 24.x / 26.x | Not known exploited — no PoC reported | No change in-window. CVE-2026-123863, CVE-2026-123865, CVE-2026-123866, CVE-2026-123872, CVE-2026-123874, CVE-2026-126051 (HTTP/2 and TLS memory disclosure / DoS / use-after-free). Six Medium and three Low complete the 16-CVE release. Node.js states 16 CVEs; one aggregator lists 17 IDs — report Node.js’s count. |
| CVE-2026-126263 — carried | 9.8 | FortiMail 8.0.0–8.0.1, 7.6.0–7.6.4, 7.4.0–7.4.6 | Not known exploited — no PoC reported | No change in-window. SQL injection (CWE-89): an authenticated attacker can execute unauthorised code/commands via crafted requests. Fixed in 8.0.2 / 7.6.7 / 7.4.7; FortiMail Cloud already remediated. Internally discovered with the Frontier AI model Claude Mythos. |
| CVE-2026-126155 — carried | 9.1 | FortiMail 8.0.0–8.0.1, 7.6.0–7.6.6, 7.4.0–7.4.8, 7.2 all versions | Not known exploited — no PoC reported | No change in-window. Incorrect authorisation (CWE-863): an unauthenticated remote attacker can obtain an authenticated IBE webmail session as any IBE recipient and read or send their encrypted mail. Fixed in 8.0.2 / 7.6.7 / 7.4.9; 7.2 must migrate to a fixed release; FortiMail Cloud remediated 5 Oct. |
| VSA-2026-062 (no CVE ID yet) — carried | — no CVSS assigned | Xen Orchestra / Xen Orchestra Appliance | Actively exploited | No change in-window. Empty-object login token gives Admin privilege; full XCP-ng host takeover as root. Fixed in Xen Orchestra 6.9.1 / 6.10.0. IOCs: 91.211.251.112:8001 and Tor domain ustar3rqxgnskzgnwkgpofgtvxevm6cx3lxdbwt4mndtveqyaohid.onion:3333. |
| SAP November Patch Day — four CVSS 10.0 — carried | 10.0 | SAP NetWeaver AS ABAP / Web Dispatcher / Gateway | Not known exploited | No change in-window. CVE-2026-76964 (ICM memory corruption, unauthenticated RCE — “ICM2SHELL”), CVE-2026-89126 (Web Dispatcher memory corruption, unauthenticated RCE), CVE-2026-89103 (Gateway broken access control), CVE-2026-120017 (SAML2 missing authentication). Patch internet-facing systems first. |
| Fortinet batch — eight criticals — STATUS, public PoC for CVE-2026-126170 | 9.8–9.0 | FortiOS / FortiManager / FortiAnalyzer / FortiSASE | Public PoC — exploitation not confirmed | New in-window development: a public proof of concept for the worst flaw, CVE-2026-126170 (9.8, JSON-RPC API, unauthenticated remote code execution on FortiManager/FortiAnalyzer), has been verified on GitHub (Ghost-Bx1/cve-2026-126170-fortimanager-jsonrpc-rce, created 11 Nov, contains working exploit.py and poc.py — genuine, not a lure); the NVD record was modified 14 Nov. Conflict — flag: PatchCVE states public exploitation activity has been reported, while Fortinet’s advisory says exploitation was not known at publication and cve.now says “public exploit code, no confirmed attacks”; the CVE is not in CISA KEV. Do not present exploitation as confirmed. The JSON-RPC API is disabled by default; the workaround is set rpc-permit none. Fixed in FortiOS 7.2.14/7.4.13/7.6.8/8.0.2 and FortiManager/FortiAnalyzer 7.2.13/7.4.12/7.6.8/8.0.2 — some fixes still “upcoming”; apply the FortiGuard virtual patch, keep OFTP/FGFM ports off the internet. |
| Microsoft November Patch Tuesday — carried | Up to 9.9 | Microsoft enterprise stack (Exchange, SharePoint, SQL Server, AD, LDAP, DNS) | One exploited (Outlook CVE-2026-124194 — see row above) | No change in-window. 1,094 newly released CVEs per Microsoft (Qualys counts 1,252 for the month; Tenable counts 1,093 — counting-method differences, flagged not resolved). Notable criticals, none reported exploited: SharePoint CVE-2026-120002 (9.9); SQL Server CVE-2026-116945 (9.8); Windows Search CVE-2026-116928 (9.8, zero-click); LDAP CVE-2026-119958 (9.8, zero-click, unauthenticated); Windows DNS CVE-2026-88218 (9.8); iSCSI CVE-2026-88234 (9.8); Exchange CVE-2026-118207 (8.8); AD DS CVE-2026-118146 (8.8); Schannel CVE-2026-119952 (8.1). Tenable notes the exploited Outlook CVE was discovered by a researcher using Claude. |
| CVE-2026-88771 — carried | 9.5 / 9.8* | Citrix NetScaler ADC and NetScaler Gateway | In the Wild (KEV) | No change in-window. Public Metasploit module (6.5.9, 6 Nov). Unauthenticated pre-auth command injection as root via attacker-controlled text written to appliance logs, later executed by maintenance script ns_monuploadd_err.pl — execution can lag the request by up to 24 hours. In KEV since 27 Sept. Fixed in 14.1-73.37 and 13.1-64.23. Patching does not remove existing implants. *CVSS v4.0 9.5 (Citrix) vs v3.1 9.8 (NVD). |
| CVE-2026-76443 — carried | 9.8 | Cisco Secure Email Gateway and Secure Email and Web Manager, 15.5 and earlier | In the Wild (KEV, sibling) | No change in-window. Public Metasploit module (6.5.9, 6 Nov). Rapid7 describes the module as an S/MIME/email-parsing SQL injection leading to unauthenticated root RCE — treat the CVE mapping as Rapid7’s, not Cisco’s. The actively exploited zero-day in the same family is sibling CVE-2026-76461 (9.8, in KEV since 14 Sept). |
| CVE-2026-97229 — carried | 10.0 | GitLab CE / EE | In the Wild (KEV) | No change in-window. Public Metasploit module (6.5.9, 6 Nov). CI/CD RegExp integer-overflow RCE as the git user; authenticated, requiring write access to a public project. In KEV since 29 Oct. |
| CVE-2026-119977 — carried | 8.1 / Low* | WordPress core 4.7.0–7.1.3 | In the Wild (attempts) | No change in-window. 240,000+ blocked attempts (peaking at 74,000+ on 30 Oct). Still no victim publicly named and no confirmed compromise. Fix 7.1.4 / backport; block web access to wp-content/upgrade/ before updating. Public PoC: lysophavin18/CVE-2026-119977 (GitHub). A second repo claiming to bypass the 7.1.4 fix holds no code, only a download link — flagged as a likely malware lure, do not download. *Wordfence 8.1 vs WordPress Low. |
| CVE-2026-107441 — carried | 8.8 (Zyxel) / 9.8* | Zyxel GS1900 series switches, ten models, web management interface | In the Wild (zero-day) | No change in-window. June fix added a check that ran after the unsafe copy in /cgi-bin/dispatcher.cgi, so June-patched firmware stayed exploitable. Payloads: Jitterly backdoor and Sixzut rootkit; suspected Chinese-speaking actor Kapibala/Red Heron (attribution via Binarly/Acronis, not independently verified). *Secondary analysis rates it 9.8. Not in KEV as of 7 Nov. Fix: 22 Oct firmware; change default credentials. |
| CVE-2026-109593 / 109594 — carried | 9.1 / 6.8 | SonicWall Global Management System 9.5.2 and earlier (end-of-life) | Not reported exploited in-window | No change in-window. 109593 critical (9.1): unauthenticated information disclosure via internal file reads. 109594 medium (6.8): high-privilege SQL injection. SonicWall will not patch — migrate to Network Security Manager. |
| CVE-2026-72898 — carried | 10.0 | Metabase 58+ (free and paid editions) | In the Wild | No change in-window. CSA Singapore warned 4 Nov that this SQL-injection flaw, fixed 6 Aug and on KEV since 11 Aug, is actively exploited and a PoC is public. Public PoC: 4minx/CVE-2026-72898 (GitHub), referenced in CSA alert AL-2026-158. Safe releases: 58.24, 59.21, 60.17, 61.11, 62.9, 63.5+. |
| CVE-2026-40217 — carried | 8.8 | LiteLLM proxy | In the Wild (LLMjacking) | No change in-window. User-ID rebind to PROXY_ADMIN. Update LiteLLM, lock endpoints off the internet, rotate keys. |
| CVE-2026-98432 — carried | 9.8 | Linux kernel IPv4 TCP, 4.4+ | Research exploit only | No change in-window. Demonstrated exploit needs MPTCP; RHEL fixed kernels require reboot, Ubuntu was still in -proposed on 3 Nov. |
| CVE-2026-58115 — carried | 10.0 | Node-RED / Siemens IoT2050 | In the Wild | No change in-window. Port 1880 without adminAuth remains an active target. Siemens scope revision noted 8 Nov (more SIMATIC IoT2050 models). |
| CVE-2026-111497 — carried | 9.2 | Google Brotli; Cisco and Splunk products (Node2Shell) | In the Wild (KEV) | No change in-window. Scope as of 5 Nov: 19 Cisco products affected, dozens under investigation, about a dozen Splunk products also affected. Same Brotli root cause also affects ClickHouse. KEV federal due date fell 10 November, but per the KEV rule a deadline alone is not a new development. Qualys published detection/verification guidance for PHP2Shell and Node2Shell on 12 Nov (four non-intrusive QIDs, two intrusive verification QIDs) and reports 167 million PHP installations across 10,000+ customers; attackers probed PHP2Shell four days before a patch existed. |
| Citrix NetScaler November batch — updated | 9.5 (highest) | NetScaler ADC, NetScaler Gateway, NetScaler Console, Secure Access Client (Windows/Linux) | Not known exploited — no PoC reported | Status — the CVE records for the batch were published 13 Nov; no new development in this window. SecurityWeek counts 28 fixes, nine critical; SecurityOnline said four bulletins fixing 31 CVEs; The Hacker News said 12 flaws. The count is still unresolved between sources — do not pick one. Headline flaws: CVE-2026-107412 (9.5, memory flaw); CVE-2026-120868 (9.3, memory overflow in packet processing, affects all ADC/Gateway deployments); CVE-2026-120869 (9.4, improper authentication in NetScaler Console, unauthenticated bypass of authentication to access restricted resources — adjacent-network, needs Console IP access); CVE-2026-118268 (8.5, Secure Access Client for Windows privilege escalation to SYSTEM) and CVE-2026-118269 (Linux equivalent). Fixes: ADC/Gateway 14.1-73.57 and 13.1-64.39; Console 14.1-74.34. Citrix says it is not aware of any exploits for the announced vulnerabilities. Separately, CitrixBleed 2 (CVE-2025-5777) is still actively exploited — OffSeq honeypots recorded 316,950 exploitation attempts from 19 source IPs in the last 30 days, last seen 13 Nov 21:41 UTC (14 Nov 05:41 SGT), with DragonForce ransomware as the severe outcome in past cases. CVE-2026-107410 (9.3, heap overflow in NetScaler Push leading to denial of service; precondition is a PUSH load-balancing virtual server) was never named in any earlier edition. |
Showing all 39 watchlist rows.
Vendor, SaaS and service-provider exposure affecting financial organisations.
Barracuda confirmed on 16 November that an attacker entered about 10 of its cloud accounts over 44 hours on 1–2 November using a cloud access key it had issued to a third-party AI company for an earlier project and never retired. The key was stored in a configuration file in the vendor’s GitHub repository and was apparently stolen when the vendor’s development environment was compromised in October. The attacker moved through a data-analytics environment, source-code repositories and Kubernetes clusters; a second attempt to create a new cloud administrator identity was blocked. Barracuda found no direct evidence that XDR customer integration credentials were viewed, but has told some customers to reissue them as a precaution.
The failure was key lifecycle with an AI-vendor wrapper, not an AI attack: the key should have died with the project. Firms should audit every key issued to a third party for finished work, and check their own GitHub repositories and config files for live credentials, because that is where this key was sitting.
Infoblox confirmed on 16 November that attackers used a compromised third-party data-migration vendor’s trusted access to exfiltrate data from its Microsoft SharePoint. The data included configuration information for some customers’ NIOS deployments — DNS, DHCP, Grid databases and reporting logs — and employee data including direct-deposit details, contact information, dates of birth and Social Security numbers. Infoblox activated incident response, hired outside experts, terminated work with the vendor and notified affected parties; it says no Infoblox products, cloud services or build processes were involved.
A migration vendor is a temporary insider with broad access, and the exposure here is the customers’ network-identity configuration — the kind of data that makes a follow-on attack against those customers much easier. The vendor is unnamed, so customers cannot check their own exposure to it; the notification to affected parties is the thing to watch for.
SecurityWeek reported on 16 November that Uber is notifying about 150,000 drivers (149,346 per a Texas Attorney General filing) after hackers used compromised service-provider credentials on 7 October to access data Uber shared with the provider; Uber learned of it on 9 October and severed connections. The provider was not named in Uber’s filings, but a spokesperson confirmed to TechCrunch it was Fireworks AI, the cloud platform for open-source AI models. Fireworks initially said no customer personal data, models or datasets were accessed; on 9 November it acknowledged account and billing information was also affected.
This is the third firm hit through the same Fireworks compromise — after Cursor and Notion — and each one learned of its exposure from the vendor, not from its own monitoring. A breach notice from an AI-platform vendor is your own exposure event, and the vendor’s first statement understated it.
SecurityWeek reported on 16 November that the Cursor supply-chain attack started at Fireworks: stolen Fireworks credentials were used to push two trojanised Claude Code extension versions (2.1.312 and 2.1.313) via Cursor’s marketplace for about 11 hours from 29 October; the attack is contained, with no attacker activity since 1 November, and Anthropic was not hacked. This answers the open question from the 16 November edition — it is now public which Fireworks intrusion exposed the key. The underlying facts are unchanged: an access key Cursor issued to Fireworks was exposed, a read-only bucket of historical coding-agent traces was copied, a small number of unredacted credentials were found inside, and up to 5,000 users may have received the malicious build. Cursor moved all inference off Fireworks, revoked the key and rebuilt the marketplace.
The cascade is now fully traced — AI-platform breach, to vendor key, to developer-tooling compromise, to marketplace poisoning. Banks using AI coding assistants should still verify no installation runs the poisoned extension builds 2.1.312 or 2.1.313, and treat agent-trace stores as credential stores.
Coinone’s breach ran through MetaM, the contractor that runs its customer support. The new detail in-window is the disclosure delay — about 70 hours 40 minutes from being told to the public notice — and that the victim count is still undetermined because unstructured chat and voice consultation data must be analysed. A fellow MetaM client, Bunjang, says only 40 phone numbers were exposed from its side. The FSS will hold a CISO meeting for exchanges and contractors on 19 November. Passwords, resident registration numbers, bank account numbers, wallet addresses and transaction IDs given during support contacts may have leaked, while Coinone says its own account, trading and wallet systems were not directly breached.
The pattern to check is which outsourced systems hold customer credentials and identity documents — a support desk is a credential store, and it was outside the firm’s own perimeter. The leaked wallet addresses plus transaction IDs also enable identity-linked balance tracing, which is a new impersonation angle.
Hotfix 3 has shipped for N-central (see the actors section), and ScreenConnect CVE-2026-84869 was confirmed exploited against MSP technicians on 12 November — a guest session compromised a technician workstation in about nine seconds. Both RMM stories remain supply-chain stories first: a compromised MSP panel or a compromised technician reaches straight into customer networks. Firms should ask their providers whether they run ScreenConnect with the RMM+ extension and whether technician clients are on 26.6.5 or later, whether they run N-central on Hotfix 3, and hunt for the RelayTrap task, the TrapWatch task, the update_<8 hex>.bat pushes and the N-central IOCs in the Indicators section.
Osaka City’s disclosure of 1,078,713 email addresses leaked via the Arara email-delivery platform was covered in the 15 November edition; no new development appeared in this window. The platform’s 6 Nov third report put the total at 81,329,467 email addresses (deduplicated) across up to about 380 client companies; another source cites about 77 million — the two totals conflict, do not merge them. The same platform was abused 28–30 Oct to send up to about 7.13 million phishing emails impersonating a credit-card company.
The open database that exposed 125 million+ SMS messages, including Swedbank account details and one-time codes, has no new development in this window. Swedbank has stopped using the provider and opened an investigation; CERT.LV is investigating. No evidence anyone other than researchers read the data; no confirmed account takeovers or financial loss.
Account takeover, mobile banking trojans, infostealers, DDoS extortion and vishing/smishing.
Microsoft Threat Intelligence published on 16 November its account of the high-volume phishing campaign that exploited the 12 November Microsoft 365 outage, observed from 13 November. The lures impersonate Microsoft 365 service-health notifications: subject pattern “Outage – <random token>”, display names “MSF Outage Notification” and “AMZ Service Health”, claiming held emails would be deleted within 24 hours unless the recipient clicked “RESTORE MESSAGES”. Sender domains are moments-notice.com, planetmaison.com and newbooker.com; click-tracking led to reallifeaccounting[.]com/themelist0.html, with outagefixed[.]sbs and outagerestored[.]sbs in some samples. Microsoft assesses it was likely credential phishing, but did not observe the final page, credential fields, submission, malware delivery or post-compromise activity, and has not attributed it to a named actor. No malware payload was observed. MITRE: T1566.002, T1036, T1204.001.
This is the vendor’s own confirmation of the campaign the 16 November edition carried from re-reports — the lure, domains and link chain are now confirmed, but the credential-theft endpoint is still unobserved, so no completed compromise is established. Credential phishing against M365 mailboxes is a classic precursor to business email compromise and cloud-data access in banks and insurers. BleepingComputer separately reported on 16 November that the outage itself was caused by a months-old networking-gateway bug, not by an attack — the final report is due 20 November.
BleepingComputer reported on 16 November that Fandango Media is notifying customers that card-skimmer code injected into its website checkout pages intercepted payment-card data for certain transactions between 8 and 11 October 2026. Confirmed exposed: card number, CVV / security code, expiry date, billing ZIP or postal code and email address. Names, Social Security numbers, government IDs and other account information were not exposed; the issue is fixed and Fandango is unaware of fraud using the data so far. The number of affected customers was not disclosed. Two years’ IDX credit and dark-web monitoring is offered, enrolment by 13 February 2027.
A Magecart-type skimmer with card data but no names is still directly usable for card fraud — the missing names slow, but do not stop, misuse. The October window means the data has been out for over a month before notification; fraud teams should watch for card-testing against the affected BINs.
Outlook India reported on 16 November that Delhi Police say hackers exploited software vulnerabilities in NBFC front-end systems over the past three months to divert approved-but-undisbursed loans into mule accounts. Twelve NBFCs in Delhi-NCR lost over ₹65 crore; 20 people were arrested and ₹18 crore was seized before withdrawal or conversion to crypto. Police cite more than 150 NBFC complaints nationwide with losses exceeding ₹300 crore in three months, using malware, cloud servers and AI tools. Hero FinCorp was among the first to report. The specific software flaws were not named.
This is a police-reported fraud pattern, not a single named-victim breach disclosure — the scale (₹300 crore-plus across 150 complaints) is the police’s figure, and the “AI tools” line is the police’s characterisation, not a technical finding. Front-end lending systems that let an attacker redirect a disbursement are a payments-control failure, and mule accounts plus crypto conversion are the laundering path.
The National Bank of Kyrgyzstan denied on 16 November (22:03 SGT) rumours of banking-system disruption circulating on social media and messaging apps, saying banks are operating normally and advising verification only via official channels. DimSum Daily likewise summarised on 16 November that no credible authority has confirmed any such threat. The rumour traces to an anonymous X post from about 31 October; Ukraine’s CCD debunked it on 9 November and Indonesia’s OJK gave an assurance on 11 November, both outside the window. The only real 17 November event is the routine US FEMA / FCC Emergency Alert System test (14:20 ET, radio and TV only) — no internet or banking impact.
No attack is happening, but the scare is itself a fraud vector: an Indonesian fact-check notes scammers are exploiting the fear with fake links, OTP requests and “move your money to a safe account” lures. Fraud teams should brief frontline staff on the hoax and its lures rather than on a non-existent attack.
JPMorgan Chase confirmed on 16 November that a “temporary issue related to a software bug in on-premise hardware systems” slowed transactions, with “no evidence this is cyber-related” and “not a systemic issue” — other banks were affected only to the extent they were party to a transaction with Chase. Customers retained access to accounts and funds. Downdetector Chase reports peaked at about 4,800 around 2pm ET (03:00 SGT 17 Nov); spikes also appeared for Bank of America, Wells Fargo, U.S. Bank, Capital One, Truist and Zelle, but BofA, U.S. Bank, Citi and Capital One said their own systems were not directly impacted, and Zelle said its network was “operating as it should”. By about 7pm ET Chase said transactions were “being processed normally now”.
This is an operational-resilience and payments signal, not a cyber incident. A single bank’s software bug slowed transactions at counterparties, which is exactly the kind of concentration risk payment operations should map.
Coinone’s leaked passwords, resident registration numbers, bank account numbers, wallet addresses and transaction IDs are now voice-phishing and account-takeover material — expect calls and messages purporting to come from Coinone, and customers should check registered withdrawal addresses; the leaked transaction IDs also enable identity-linked balance tracing. The FSI special audit began on 16 November — expect phishing citing the FSI or the evaluation process, using the real assessment-staff contact details and firm names from the leak. The SBI FIR names a group, “Siriusim”, and confirms loan-application and account-opening data was in the stolen set — expect phishing citing SBI loan and account-opening details. Monex’s 1.13 million email addresses and AXA’s about 597,000 accounts remain impersonation material, as do KOMSCO’s chak-user details (KOMSCO warned of voice-phishing and smishing; a self-check function opens in the chak app from 17 November) and KBIZ’s Yellow Umbrella details (KBIZ warned of impersonation and loan fraud). Arara’s leaked shareholder emails are live phishing bait, and the same platform was already abused to send about 7.13 million phishing emails impersonating a credit-card company. The Tesouro Nacional leak (names, CPFs, Pix keys) and Caixa’s leaked CPFs are impersonation material for 7.6 million-plus people. SMBC Nikko’s leaked holdings and account details are impersonation material for shareholders. Mirae Asset’s leak gives attackers 3,800 customer account numbers plus names, emails and mobiles of 35,255 employees. Binance, Mercado Bitcoin, Gopax and Nubank/Mercado Pago outage messages remain live phishing bait, and the Microsoft 365 outage credential-phishing campaign (sender domains moments-notice.com, planetmaison.com, newbooker.com; click-tracking reallifeaccounting[.]com) is now confirmed by Microsoft — verify via official status pages only. Autopay customers in Poland should watch for phishing citing the payment processor, after customer notifications named suspected access to email, phone and payment-card data. Watch for Rs 1 test transactions on compromised accounts.
N0n’s leak-site listings posted on 16 November claim cardmember programme data from Visa, Mastercard and JPMorgan Chase, plus client records from D.R. Johnson Financial and Tailored Benefits. None of the firms has confirmed a breach, no data samples have been published, and Bitdefender has shown N0n inflates its victim count. Fraud teams should not treat any of the five as a confirmed breach, but should watch for phishing citing cardmember programme, travel-booking and benefits-plan details, and for any company statement or data sample.
Customer allegations of thefts from Bank of Khartoum accounts via the Bankak mobile banking app remain reported / unconfirmed: more than 50 traders and customers in West Kordofan claim millions of Sudanese pounds were stolen, with individual reported losses of SDG 18–45 million. The bank has still not responded to comment requests, and no technical evidence that the app itself was breached has been published. Treat as an allegation, not a confirmed bank breach.
No new DDoS-extortion item targeting a financial institution was found in this window. The CSC Telecom open database (125 million+ SMS messages including Swedbank one-time codes) was published on 11 November, outside the window, and has no new development.
Techniques actually published for this edition’s actors and key vulnerabilities. No mapping is shown where the source published none. The FBI FLASH-20261113-001 mapping for ARTEX was carried in the 16 November edition and is not a new sighting in this edition.
| MITRE technique ID | Technique / behaviour | Seen in this edition | BFSI relevance / detection note | Status |
|---|---|---|---|---|
| T1566.002, T1036, T1204.001 | Phishing: spearphishing link; masquerading; user execution | Microsoft 365 outage phishing campaign — Microsoft Threat Intelligence blog, 16 Nov | Outage-themed service-health lures with subject “Outage – <random token>”, display names “MSF Outage Notification” / “AMZ Service Health”, threatening deletion of held emails within 24 hours. Sender domains moments-notice.com, planetmaison.com, newbooker.com; click-tracking reallifeaccounting[.]com/themelist0.html. Detect on the sender domains and the outage subject pattern; verify via official status pages only. | Confirmed |
| — none published | Cloud access key theft and use: unretired vendor key, stored in a configuration file in a public code repository, used to enter cloud accounts (Barracuda) | Barracuda Trust Center advisory (10 Nov, updated 13 Nov); BleepingComputer, 16 Nov | No MITRE IDs were published for the Barracuda incident. Detection rests on auditing keys issued to third parties for finished projects and on scanning code repositories and configuration files for live credentials. The ARTEX pattern-match is the vendor’s assessment, not a published technique list for this intrusion. | No mapping published |
| — none published | Trusted third-party access abused to exfiltrate from a collaboration platform (Infoblox / SharePoint migration vendor) | Infoblox, 16 Nov; The Register, 16 Nov | No MITRE IDs were published. Detection rests on reviewing third-party access to collaboration systems and on watching for notifications about vendor-held configuration data. Customer NIOS configuration data in attacker hands is follow-on attack material. | No mapping published |
| — none published | Unauthenticated remote code execution via Apache Axis unsafe reflection (CVE-2026-90436), exploited against Confluence Data Center and N-central | Rapid7 Emergent Threat Response, 16 Nov | No MITRE IDs were published in the Rapid7 post. Detection rests on the published IOCs and on checking whether Confluence’s Remote API (XML-RPC & SOAP) or N-central’s Legacy API is enabled — both are non-default preconditions. | No mapping published |
| — none published | Guest-session file-transfer abuse in ScreenConnect (CVE-2026-84869): guest pushes a batch file the technician’s host client runs without authorisation | Huntress, 16 Nov | No MITRE IDs were published in the Huntress post. Detection rests on guest “run file” requests for update.bat / update_<8 hex>.bat and on the masterdarky[.]com/WindowsUpdate.msi download. The fix sits on the technician client (26.6.5+), not the server. | No mapping published |
| — none published | API-authentication flaw exploited to read customer records (KOMSCO chak platform) | KOMSCO second statement (16 Nov; Korean press — translated) | No MITRE IDs were published. Detection rests on API-authentication logging on inter-agency interfaces. The same weakness class — an authentication flow, not a web server — also hit the Korean FSI. | No mapping published |
| — none published | N-able 4EverAble2Shell / Agent2Shell chains (HF3 shipped 16 Nov) | N-able status, 16 Nov; Rapid7, 16 Nov | RMM used by MSPs serving small banks. The Legacy API is now permanently blocked by HF3. No new vendor-published mapping in this window; the secondary-summary IDs (T1190, T1072, T1136, T1552, T1496, T1068) were published on 13 Nov, not in this window, and are not carried forward as a new sighting. | No mapping published |
| — context only | ARTEX agentic intrusion chain (recon through exfiltration inside hours) | Google Threat Intelligence Group AI Threat Tracker, 16 Nov; Unit 42 CL-STA-1238 (via Dark Reading, 16 Nov) | No new MITRE IDs were published for ARTEX in this window. GTIG confirms agentic frameworks in live intrusions and a fully automated patch-to-exploit pipeline (exploit_agent), with no success rate or victim count published. Unit 42 reports a second documented Cairn campaign (25 operations, Aug–Sep, 10 organisations breached; no BFSI victims named). The FBI FLASH mapping from 13 November remains the standing reference and is not repeated as a new sighting. | Context only |
Recurring-TTP watch: the technique that keeps recurring in this edition is trusted third-party access — an unretired vendor key (Barracuda), a migration vendor’s SharePoint access (Infoblox), an AI-platform compromise reaching customers (Fireworks / Uber, Cursor), and a support contractor’s server (Coinone / MetaM). Alongside it, the same older patterns persist: exploitation of internet-facing management and edge interfaces (RMM panels, gateways, VPNs), weak or default credentials on exposed services, and outage-themed credential phishing timed to degraded defences. What is new is the compression GTIG describes: a fully automated patch-to-exploit pipeline and agentic frameworks chaining the whole intrusion inside hours. No technique was published with a MITRE ID for Barracuda, Infoblox, KOMSCO, the N-able chains or the ScreenConnect exploitation in this window — that gap is stated, not filled.
The three AI roles are kept separate. Vendor forecasts are marked as forecasts.
Google Threat Intelligence Group — LLMJacking hit financial services (16 Nov). GTIG’s AI Threat Tracker, published 16 November at 18:30 UTC (17 Nov 02:30 SGT), says Mandiant responded to several LLMJacking incidents in 2026: financially motivated actors hijacking cloud and LLM credentials and quotas, with single-day fraudulent charges over $100,000 and worst cases reaching $1 million. The November LLMJacking write-up covers intrusions affecting financial services, retail, transportation and technology organisations in North America, South America and Asia-Pacific, including abuse of LiteLLM CVE-2026-40217 to escalate virtual keys to admin. GTIG also tracks “LLM Mixers” reselling stolen LLM access at $0.10–$25 per million tokens.
Barracuda — the target was an AI vendor’s development environment (16 Nov). The key used in the Barracuda intrusion was stolen when a third-party AI company’s development environment was compromised in October; the key then sat in a configuration file in the vendor’s GitHub repository. The AI vendor is unnamed.
LLMJacking is now a confirmed financial-services problem, not a hypothetical — the charges are the measurable damage, and the LiteLLM flaw in the watchlist is the same product class being harvested in practice. Banks self-hosting AI gateways or inference endpoints should treat them as internet-facing credential stores, not internal tools, and should rotate any key that was ever exposed. The “LLM Mixer” market means a stolen key is monetised quickly, so detection time matters more than the key’s nominal value.
GTIG AI Threat Tracker — agentic attacks cross a threshold (16 Nov). Google’s quarterly tracker says adversarial AI has reached a “critical inflection point”: actors have moved beyond manual prompting to autonomous multi-agent frameworks in live intrusions. Confirmed or observed in the wild: ARTEX, CAIRN (against US aerospace targets), CyberStrikeAI and PentAGI (a customised PentAGI used from late August to early October 2026 for reconnaissance and scanning across Asia), with experimentation on HexStrike AI. For the first time, GTIG reports a fully automated vulnerability-discovery-to-exploitation pipeline — a framework it calls exploit_agent, found on an exposed staging server, which diffs patches, drafts exploits and executes them with no human in the loop (Russian-language configuration comments; victim telemetry suggested interest in French media / entertainment and education sites). GTIG gives no success rate or victim count — do not present one. Supporting speed evidence: LLM-built working exploits for chained PaperCut zero-days CVE-2026-81578 / CVE-2026-82078 within a day of disclosure in late August, and a GreyNoise-reported widespread campaign four days later assessed as likely LLM-built; zero-day exploits spiked to 22 in August 2026 against an 11-per-month 2026 average. ExtraHop separately counted 3,447 ARTEX hosts on 16 November (reported via CyberPress).
Unit 42 — China-nexus CL-STA-1238 used Cairn to breach 10 organisations (16 Nov, via Dark Reading). China-nexus actors ran 25 operations in August–September against at least 19 organisations, achieving unauthorised access or data retrieval at 10. Seven targets were semiconductor / advanced-manufacturing supply chain, three US defence / defence-policy, plus human-rights and civil-society groups. Impact included dozens of valid credentials, more than 1 million records exposed at one organisation, military and government identities retrieved, persistent admin accounts created, and a live industry standards-ballot document modified. Cairn (model-agnostic; here driven by Zhipu GLM 5.2 and DeepSeek V4 Flash / Pro) planned, executed and recorded operations autonomously from a target and a goal. Initial access ran through known weaknesses — authentication bypasses, IDOR, mass-assignment privilege escalation, weak or default credentials, exposed cloud APIs and dated public CVEs. No BFSI victims were named.
Suspected DPRK cluster MIDNIGHT NEPTUNE targeted crypto organisations (GTIG, 16 Nov — suspected, not proven). GTIG reports a suspected DPRK cluster targeted cryptocurrency organisations in North America and Europe via fake-recruiter personas (SynthID-watermarked imagery) and ClickFix lures, then modified .claude/settings.json in internal repositories so launching Claude Code downloaded the SOMBERMEME backdoor, masking movement with GitHub personal access tokens named after Microsoft Copilot.
KOMSCO “AI-used attack” and Autopay “LLM-assisted attack” — company characterisations only. Segye reports KOMSCO characterised the chak breach as an “AI-used attack, similar in pattern to recent financial hacks”, with no technical evidence published. Autopay’s security manager described its attack as highly automated and likely LLM-assisted, with no malware or phishing — again with no technical evidence published. Neither is presented here as a confirmed AI-driven attack.
No new AI-scaled phishing, vishing or deepfake campaign against a named bank was found in this window.
CrowdStrike — backdoored LoRA models are a supply-chain risk (16 Nov; lab research, not in-the-wild exploitation). CrowdStrike’s Cyber Superintelligence Lab built on ProjectDiscovery’s October 2026 proof of concept (a roughly $50 LoRA fine-tune of Qwen2.5-7B trained to emit a tool call to an attacker URL on a trigger phrase) and built a two-stage pre-deployment defence: static weight analysis to flag undisclosed merged LoRAs, then trigger recovery to confirm the backdoor and recover the activating input. Across 16 lab models (8 backdoored, 8 benign), the scanner flagged all 8 backdoored models and Stage 2 confirmed a trigger in every one — covering tool-call hijacking, injectable shell code, typosquatted imports (for example pandas2) and refusal-flips — and cleared the 2 benign models escalated from Stage 1. A side finding: fine-tuning alone can erode refusal behaviour (“alignment erosion”). Limits stated: 16 models is a test, not a population; full fine-tunes and direct weight edits are not addressed. CrowdStrike states no in-the-wild exploitation is reported.
Barracuda key lifecycle is a governance failure, not a model failure. The key that let an attacker into Barracuda’s cloud was issued to a third-party AI company for an earlier project and never retired, then stored in a configuration file in the vendor’s GitHub repository. The control problem is vendor-key inventory and retirement, plus scanning code repositories for live credentials — the same discipline the Cursor / Fireworks cascade demands.
Crédit Mutuel Arkéa — chairman’s AI claim is still uncorroborated. An aggregator piece dated 16 November (03:55 SGT 17 Nov, citing el-balad) repeats chairman Julien Carmona’s claim that advanced AI was very likely involved in the attack and that about €100m was drained via a third-party provider. No primary French source was found for the AI remark, and no supporting evidence was given. Do not present the AI claim as confirmed. The $108.8m Bitquery figure conflicts with the earlier estimate of more than $133m (ZachXBT) — do not merge the two figures.
GTIG’s “LLM Mixers” and the patch-to-exploit pipeline compress the control window. Stolen LLM access is resold at $0.10–$25 per million tokens, and a fully automated pipeline now diffs patches and drafts exploits with no human in the loop. For BFSI, the practical consequence is that the time between a patch landing and an exploit existing has shrunk to hours — patch windows and key-rotation windows need to match that speed, not a monthly cycle.
Notable indicators from this edition, and prioritised steps for SOC and vulnerability-management teams.
Every source is hyperlinked to the original vendor, government or primary reporting; translated items link to the original-language article.