BFSI Cyber Threat Intelligence Briefing

Wednesday 4 November 2026

Window: approximately 3 November 09:00 SGT to 4 November 09:00 SGT. New and overnight developments, deduplicated against the 3 November edition, with gaps stated plainly.

Window 24 hours to ~09:00 SGTKEV total 1,767 entriesDue this week WordPress 6 Nov · Brotli 10 Nov · VMware 11 Nov · Ivanti 13 Nov
Due Friday 6 Nov
WordPress KEV

CVE-2026-119977 — patch and review upgrade-window logs.

Due 3 Nov — closure check
Siemens & Plex

Confirm V4.0.9.8 and 1.43.3.10861+; hunt, do not just check versions.

Supply chain
cPGuard compromise

Versions 5.88.12–5.88.19 shipped a backdoor; rotate credentials.

Exchange watch
Binance / Mercado

Binance 98% replaced; Mercado still suspended, no reopening date.

Executive summary

1

The one new KEV entry is WordPress, due Friday 6 November — and the score itself is disputed. CISA added CVE-2026-119977 on 3 November, taking the catalog from 1,766 to 1,767 entries, with forensic triage requested. Wordfence rates it 8.1; the WordPress project rates the same issue Low. Exploitation attempts have been seen since 28 October, but no successful compromise via this race has been publicly confirmed in sources read. Patch to 7.1.4 or the branch backport, and check logs.

2

Two supply-chain and edge stories moved from warning to action on 3 November. OPSSHIELD disclosed that cPGuard versions 5.88.12–5.88.19 shipped a backdoor and harvested password hashes, SSH files and root database configuration after its update channel was compromised on 2 November; updating to 5.88.25 or later removes the backdoor but does not un-steal credentials. Separately, Barracuda disclosed two flaws (CVE-2026-116606, 9.3, and CVE-2026-111560, 8.6) that chain to root on its Email Security Gateway — with no exploitation reported yet.

3

The exchange recovery story split in two directions. Binance said on 3 November it had replaced 98% of affected assets, with a wider proof-of-reserves snapshot due 4 November — but that proof still covers only 10 assets at a 31 October snapshot, so it is evidence about balances at a cut-off, not an audit and not a post-mortem. Mercado Bitcoin revised its own figure twice in a day (about $38m, then about $36.3m affected with about $12.9m blocked) and still gives no reopening date; blocked funds are not recovered funds.

Threat actor activity

Group Name | Target Sub-sector / Geography | TTPs (MITRE ATT&CK where available) | Impact

GroupTargetTTPsImpact
TraderTraitor — UpdateCrypto exchange, global; BinanceNo new MITRE list in-window. Binance: 98% of affected assets replaced, further withdrawals from 04:00 UTC 3 Nov, wider snapshot due 4 Nov. North Korea (KCNA, 3 Nov) denied involvement; the US reiterated its concerns the same day. Zscaler's trojanised Terraform campaign (FLATROOF / ROOFDECK) was recirculated 3 Nov but dates from 8 Oct.A denial does not change the task: the FBI wallet list stands, restoration phishing is the live risk, and 98% replenished is not recovery of stolen funds. The 10-asset proof does not disclose other liabilities or the attack path.
Unknown — Mercado Bitcoin — UpdateCrypto exchange, BrazilNo MITRE mapping. Exchange figure revised: about $38m affected / $11m blocked, then about $36.3m / $12.9m blocked on 3 Nov. Gradual reopening, no date. Entry method undisclosed.Still suspended. Researcher estimates ($40m–$49m) still exceed the exchange's figure. Blocked is not returned.
Unknown — BTCPay drains — UpdateBitcoin payment processing, globalNo new MITRE mapping. Bitquery trace reported 3 Nov: at least $1,628,450 across 682 sweeps (18.38 BTC, 3,405 LTC), 30 Oct–2 Nov, to BTCPay's published addresses plus one associated wallet. Sweeps continued after the 1 Nov warning.$1.63m is a floor, not a total. BTCPay has disclosed no official total or instance count. Pre-2.4.2 instances should still be treated as potentially compromised.
UNC6993 and other LLMjacking clusters — NEWFinancial services, tech, retail, transport; Americas and APACT1078.004, T1098.001 (secondary summaries). Stolen cloud / LLM keys or exposed endpoints resold via "LLM mixer" services such as Jet Night; 500,000+ calls in under eight hours in one case. LiteLLM CVE-2026-40217 abused for PROXY_ADMIN keys.Charges above $100,000 in a day, up to about $1m before revocation. A stolen AI key is a billable asset monetised within hours. No named bank victim.
Agentic-framework intrusions incl. UNC7228 — NEWMultiple sectors; PRC-nexus link (STORM-0940)No full MITRE list. Since late Aug 2026, ARTEX, PentAGI and CAIRN run inside Mandiant engagements. GTIG's stage-6/7 forecasts are assessments, not observations, and no engagement counts or success rates are given.Detection should look for agent-tool behaviour (rapid, tool-driven command sequences), not wait for a named malware family.
Videostrong / XCPE botnet — NEWAndroid TV boxes, globalNo MITRE mapping. Manufacturer-built firmware backdoor (Google: high confidence) enrolled 1.3m+ boxes; used for ad fraud, data theft and residential proxies. Google seized control domains and banned Videostrong.No bank victim. The 10m figure is Google's estimate of additional devices severed via shared control domains, not XCPE devices. Firmware backdoor persists after takedown.
Direwolf / INCRANSOM / Akira — UNVERIFIED listings, 3 NovUSClaims (US, financial services); fidepartners.com (Spain); SK Global Software (South Korea, banking/treasury software)No MITRE lists. Leak-site claims only; no confirmed encryption, exfiltration volume or ransom demand. Direwolf listing date differs between sources (29 Oct vs 3 Nov).Unconfirmed. SK Global is supply-chain-adjacent if verified, but none of these is a confirmed bank breach.
Unknown US bank phishing operation — NEW (published 3 Nov)Mostly midsize/regional US banks; also UK, Netherlands, Australia, NZNo MITRE mapping. 248 live lookalike domains impersonating 189 US firms as of 25 Oct, behind Cloudflare and cloaking. No domains/IPs released publicly.Credential capture usable for wire/ACH fraud. A page that looks dead on repeat visit may be cloaking, not takedown.

No new in-window item was found for FIN7 / Carbanak, Scattered Spider, RansomHub, Cl0p, or a FI-specific Initial Access Broker. Absence of reporting is not evidence of absence of activity.

Vulnerability watchlist

CVE ID | CVSS Score | Affected Product | Exploitation Status | Relevance to Finance

CVECVSSProductStatusFinance relevance
CVE-2026-119977 NEW · KEV8.1 / Low*WordPress core 4.7.0–7.1.3In the Wild (attempts)Customer/marketing sites. Due 6 Nov, forensic triage. Fix 7.1.4 / backport; block wp-content/upgrade/. *Wordfence 8.1 vs WordPress Low — same issue, different ratings.
CVE-2026-116606 NEW9.3Barracuda Email Security GatewayNot known exploitedPre-auth config-import bypass; chains to root with CVE-2026-111560. Patch ESG 9.4.0.030+; admin interface off the internet.
CVE-2026-111560 NEW8.6Barracuda ESG, ADC, Archiver, WAF, WAF Control Center, Web Security GatewayNot known exploitedAdmin command injection as root. Patch per product (see briefing text).
CVE-2026-98432 NEW update9.8Linux kernel IPv4 TCP, 4.4+Research exploit onlyRHEL fixed 1 Nov (reboot required); Ubuntu still in -proposed 3 Nov. Demonstrated exploit needs MPTCP — do not overstate as universal remote root.
CVE-2026-109618 / 109619 (PHP2Shell)9.2PHP; PAN-OS, Cisco, Citrix, F5, Siemens productsIn the WildCSA alert 3 Nov: PAN-OS management interfaces exploited, chained with CVE-2026-46300 to root. Remove management interfaces from the internet.
CVE-2026-1017679.8Oracle Application Object Library 12.2.3–12.2.15Exploitation attempts reportedCSA alert 2 Nov (missed-item correction). Apply Oracle Oct 2026 update; CVE-2026-101700 (8.1) in same update.
CVE-2026-402178.8LiteLLM proxyIn the Wild (LLMjacking)User-ID rebind to PROXY_ADMIN. Update LiteLLM, lock endpoints off the internet, rotate keys.
CVE-2025-119699.4WSO2 Open Banking AM/IAM, API ManagerNot known exploitedDirect Open Banking relevance. Vendor 3-day window for internet-facing instances falls due 5 Nov.
CVE-2026-5811510.0Node-RED / Siemens IoT2050In the WildAutomation/OT hosts; port 1880 without adminAuth remains an active target.
CVE-2026-1114979.2Google BrotliIn the Wild (KEV)Due 10 Nov. Cloud/web tier for financial workloads.
CVE-2026-110936 / 96652 / 966558.4 / 5.3Siemens Software Center; Plex Media ServerIn the Wild (KEV)Due 3 Nov — now closure checks. Siemens V4.0.9.8; Plex 1.43.3.10861+.
CVE-2026-417098.8*VMware ESX/ESXiIn the Wild (KEV)Due 11 Nov. *Per secondary reporting.
CVE-2026-1185274.3Ivanti Connect SecureIn the Wild (KEV)Due 13 Nov. CVE-2026-111357 in the same chain was due 2 Nov — rebuild suspect appliances.

No new core-banking, SWIFT/payment-gateway or enterprise file-transfer CVE was reported in this window in sources searched. KEV now stands at 1,767 entries.

Supply chain & third-party risk

cPGuard / OPSSHIELD — NEW, 3 Nov

Stolen credentials let an attacker into cPGuard's update system on 2 Nov and push versions 5.88.12–5.88.19 carrying a backdoor that harvested /etc/shadow, /etc/passwd, root database configuration, shell history and SSH files (staged as loot.tgz in /var/tmp/.cpg91511). OPSSHIELD provisionally links confirmed retrieval to 32 servers; a secondary expanded figure (~2,950 servers, 7,900 billing accounts) could not be verified against the advisory and is not adopted here. Fix to 5.88.25+ and rotate all credentials that could have been on an affected server.

Google / Videostrong XCPE — NEW

A certified maker's firmware backdoor, not an update compromise. Banning the maker and seizing control domains removes the operator's channel but leaves backdoor code in boxes already sold — treat affected hardware as untrusted, not cleaned.

Abrigo / Sageworks — Update, unchanged

No new Abrigo statement in-window: Analyst and API are back, borrower-facing Online Portal Now remains unavailable, and the breach-or-not question is still unanswered. The outstanding ask for an attestation on scope and data types stands.

Barracuda as perimeter vendor — NEW

A security appliance that can be reconfigured or rooted from its admin interface is a trusted-component failure. Patch, then remove the admin interface from internet reach.

Emerging fraud & attack vectors

MoneySmart breach confirmed 3 Nov

MoneySmart (Singapore / Hong Kong) confirmed an unauthorised third party accessed some internal systems, discovered 29 Oct, and copied personal data of some customers. Expect follow-on targeted phishing and calls quoting real product details. MoneySmart says it never asks for passwords or OTPs by phone, email or text.

HKMA warning, 3 Nov

Fraudulent websites and internet-banking login screens reported for the Bank of East Asia, Shanghai Commercial Bank, Chong Hing Bank and HSBC, among others. Banks will not send hyperlinked SMS/emails to carry out transactions or ask for passwords/OTPs.

Belize bankers' warning, 3 Nov

Ongoing phone/messaging scams impersonating bank representatives to obtain credentials and password-reset codes. Legitimate institutions never solicit confidential information that way.

Exchange restoration phishing — ongoing

Binance and Mercado Bitcoin restoration stories are live phishing bait. Verify via official status pages only. Mercado Bitcoin repeats: it never asks for a password, code or recovery phrase, and offers no refunds via links, WhatsApp or phone calls.

TTP tracker — threat actors and key vulnerabilities

Techniques actually published for this edition. No mapping is invented where a source published none, and older research recirculated in-window is marked as context, not as a new sighting.

MITRE techniqueBehaviourSeen in this editionBFSI relevance / detection
T1078.004Valid Accounts: Cloud AccountsUNC6993 / LLMjacking — stolen cloud, GitHub and LLM keys used to run inference at the victim's expenseA stolen AI or cloud key is monetisable within hours. Alert on new principals, spend spikes, and unusual key locations. Reported in secondary summary of GTIG / Mandiant research.
T1098.001Account Manipulation: Additional Cloud CredentialsUNC6993 / LLMjacking — additional cloud credentials created / abused to sustain model accessRevoke-and-rotate, do not merely cap spend. Audit who can create additional credentials.
T1059.004Command and Scripting Interpreter: Unix ShellZscaler trojanised Terraform campaign — FLATROOF / ROOFDECK, recirculated 3 NovContext only. Underlying research is dated 8 Oct, so this is not a new 3 Nov attack.
T1105Ingress Tool TransferZscaler trojanised Terraform campaign — FLATROOF / ROOFDECK, recirculated 3 NovContext only, same 8 Oct caveat as above.
No mapping publishedAgent-tool behaviour — rapid, tool-driven command sequences from a single sessionAgentic intrusions incl. UNC7228 — ARTEX, PentAGI, CAIRN since late Aug 2026Detect the behaviour, not a malware name. GTIG stage-6/7 forecast is an assessment, not an observation.
No mapping publishedPre-auth bypass, then root command injectionBarracuda CVE-2026-116606 chained with CVE-2026-111560Patch-before-exploitation, not a confirmed incident chain. Email is the phishing front door for every bank.
No mapping publishedUpdate-channel compromise, credential harvesting, backdoorcPGuard / OPSSHIELD — versions 5.88.12–5.88.19Treat any server that installed those versions as potentially compromised, even if backdoor files are gone.
No mapping publishedFirmware backdoor persisting across rebootVideostrong / XCPE — 1.3m+ Android TV boxesA control-domain takedown does not clean the device. Treat affected hardware as untrusted.
No mapping publishedLookalike-domain credential capture with cloakingUS bank phishing — 248 live domains, 189 US firms, 25 OctA page that looks dead on repeat visit may be cloaking, not takedown. Capture is a wire/ACH fraud precursor.

Recurring watch: valid-account / cloud-account abuse (T1078.004) links LLMjacking to broader cloud takeover — the same exposed endpoint or leaked key pattern recurs across actors. No other technique in this edition was published with a MITRE ID.

AI in the current threat landscape

AI as a target, AI as a weapon or accelerator, and AI as a risk and control problem for BFSI.

AI as a target — LLMjacking, live today

Stolen service-account keys, GitHub tokens or LLM API keys, or exposed endpoints (including Ollama), are used to run inference at the victim's expense and resell it via "LLM mixer" services such as Jet Night. One actor made 500,000+ calls in under eight hours; charges above $100,000 in a day, up to about $1m before revocation. LiteLLM CVE-2026-40217 (8.8) turned ordinary virtual keys into PROXY_ADMIN credentials. No bank victim is named, and the figures are reported scales, not audited bank losses.

AI as a weapon — agentic intrusions, live today

Since late August 2026, actors have run ARTEX, PentAGI and CAIRN inside Mandiant engagements, turning high-level goals into commands via model APIs. GTIG places activity at stage 5 and predicts stage 6 within 1–3 months and stage 7 within 3–6 months. That forecast is an assessment, not an observation — no engagement counts or success rates are given.

AI as a risk and control problem for BFSI

Govern: (1) API and service-account key inventory and rotation; (2) endpoint exposure — Ollama / LiteLLM must not be internet-reachable without authentication; (3) spend and invocation monitoring, with revoke-and-rotate, not just spend caps; (4) agent-tool detection — rapid, tool-driven sequences from a single session; (5) open-source tooling — blocking one framework name does not block the technique.

No new AI-scaled phishing item in this window

No newly reported AI-generated phishing, vishing or deepfake campaign against a BFSI victim was found for 3–4 November. That is absence of reporting, not evidence it has stopped. The MoneySmart, HKMA and Belize warnings in this edition are conventional impersonation and are not re-labelled as AI campaigns without evidence.

Indicators & defensive actions

Notable IOCs / behavioural indicators

Prioritised actions — SOC

  1. Deadline order: WordPress by 6 Nov (patch + log review); Brotli by 10 Nov; VMware by 11 Nov; Ivanti CVE-2026-118527 by 13 Nov.
  2. Confirm 3 Nov closures: Siemens V4.0.9.8 and Plex 1.43.3.10861+ with hunt-and-rebuild tests, not version numbers alone.
  3. Hunt cPGuard today across your estate and your hosting providers'; update to 5.88.25+ and rotate every credential that could have been on an affected server.
  4. Hunt Barracuda and Linux TCP exposure: admin interfaces off the internet; inventory MPTCP listeners and kernel versions; apply IP-options filtering where no production fix exists.
  5. Brief crypto desks and customers: Binance's 4 Nov snapshot is not a post-mortem and covers 10 assets; Mercado Bitcoin remains suspended with no date; both are phishing bait. Use banks' own channels for the Hong Kong and Belize warnings.
  6. Check AI/cloud key exposure: audit LiteLLM/Ollama endpoints, service-account keys and committed API keys; alert on spend spikes and new-principal invocations; revoke-and-rotate.

Prioritised actions — Vulnerability Management

  1. By 6 Nov: WordPress CVE-2026-119977 to 7.1.4 or branch backport.
  2. Immediately: Barracuda patches; cPGuard to 5.88.25+ with rotation; Linux kernel fixes (reboot) or IP-options filtering; LiteLLM update and key rotation; Oracle October 2026 update (CVE-2026-101767 / CVE-2026-101700).
  3. This week: Brotli by 10 Nov; VMware by 11 Nov; Ivanti by 13 Nov. WSO2 CVE-2025-11969 internet-facing instances fall due 5 Nov.
  4. Verify closure of the 3 Nov deadlines (Siemens, Plex) and the 2 Nov Ivanti deadline (rebuild suspect appliances).
  5. Third-party: demand Abrigo's breach-or-not attestation; ask hosting providers about cPGuard 5.88.12–5.88.19; ask Barracuda owners to confirm admin-interface exposure, not just patch level.

Could not verify

Mercado Bitcoin entry method, affected system and single loss total; BTCPay entry method, instance count and official total; Binance's 98% figure and reserve ratios (unaudited, no post-mortem yet); cPGuard scope (32 servers provisional vs ~2,950 secondary, unverified); the three 3 Nov ransomware listings (leak-site claims only, Direwolf date conflict); a successful WordPress compromise via CVE-2026-119977 (attempts confirmed, compromise not); the XCPE 10m figure (estimate of additional devices, not XCPE boxes); LLMjacking/ARTEX prevalence (no counts, success rates or victim names). No in-window evidence was found for new FIN7/Scattered Spider/Carbanak/RansomHub/Cl0p BFSI attacks, FI-specific IAB sales, SWIFT/core-banking/file-transfer CVEs, a new mobile-banking trojan, or a new public FS-ISAC alert.

Sources

Previous edition (deduplication baseline): 3 November 2026 briefing, in the goal workspace at briefs/2026-11-03-briefing.md. This page is the 4 November edition; previous days' pages are untouched.

← Back to main page