Weekly Threat Digest 26.46 || Nov 9
APT29 watering holes hit iOS and Android as a WebKit flaw joins KEV; OpenSSH "sshhole" bypass hits Alpine
Sekoia assesses, with moderate to high confidence, that APT29 (also tracked as Midnight Blizzard) is running watering hole attacks that serve iOS and Android exploit chains built on known, patched flaws. An OpenSSH flaw dubbed "sshhole" (CVE-2026-123158) lets attackers log in without a password on Alpine hosts that allow password logins, and a public exploit is available.
Unit 42 reports that attackers set refrigeration controllers at European supermarkets to defrost mode, and French authorities disclosed that medical facilities were affected. CrowdStrike details VAPOR PANDA's use of ARTEX AI agents against telecommunications providers and others, and Fortinet is investigating which of its products a critical Linux kernel IPv4 flaw affects.
Act Now
- APT29 mobile watering holes (WebKit): APT29 watering holes serve iOS and Android exploit chains, and CVE-2024-27834 was added to KEV Nov. 4. If you run iPhones or Android: update iOS and Chrome to current vendor releases, apply Android patches, and check proxy logs.
- OpenSSH on Alpine ("sshhole"): A long password logs an attacker into accounts with no password set on Alpine's default sshd, and a public exploit exists. If you run sshd on Alpine: update
opensshand restart sshd, or setPasswordAuthentication no.
Drafted with AI from public reporting; reviewed and edited by GPP Labs before publishing.
OT/ICS Advisories
Marker shows the highest CVSS v3 score in each advisory: 🔴 9.0 and up · 🟠 7.0–8.9 · 🟡 4.0–6.9
- 🔴 Subnet Solutions PowerSYSTEM Center (power system software) · ICSA-26-307-01 · 2 CVEs · CVSS v3.1 9.0. Used in energy and critical manufacturing environments to manage power systems. Two flaws let an attacker run unauthorized SQL commands and reach resources they should not be able to access. No patch for the CVEs; upgrade to 2026 Update 1.1 (2024 and 2026 users) or 2020 Update 30 CISA
Critical Infrastructure
Exposed toolkit attributed to DoNot names Pakistani nuclear and oil and gas targets
InfraHunter reports an open directory of 48 files uploaded Oct. 9-14, 2026, including Go reverse shells, an HTTP beacon, a Metasploit loader, and Inveigh scripts for capturing NTLM hashes. Reconnaissance scripts hardcode references to the Pakistan Atomic Energy Commission, the Chashma and KANUPP nuclear plants, SCADA and Modbus, and the state oil and gas company OGDCL, and several target an account named ELECTRICAL. The researcher attributes the kit to DoNot (APT-C-35) based on infrastructure and macro overlaps. No compromise is reported.
- Exposure: Windows networks at energy and nuclear organizations where LLMNR, NBT-NS, or mDNS name resolution is on, which the kit's Inveigh scripts abuse to capture NTLM hashes.
- Response: Disable LLMNR and NetBIOS name resolution where they aren't needed, and alert on hosts answering name-resolution requests they don't own.
- Detections: InfraHunter published IP addresses including
193.29.104[.]84and179.43.145[.]232, a domain, a file hash, and the kit's file names. - Also reported by: CyberVeille
Vulnerabilities & Exploits
APT29 watering holes hit iOS and Android with n-day exploits
Sekoia reports a large-scale watering hole campaign that serves iOS and Android exploit chains through compromised sites, lookalike domains, and a one-week compromise of the Plerdy SEO Tracker. Sekoia assesses with moderate to high confidence that APT29, also tracked as Midnight Blizzard, did this. CVE-2024-27834, a WebKit flaw used in the iOS 16.7 chain, was added to CISA's Known Exploited Vulnerabilities (KEV) catalog this week.
Injected JavaScript fingerprints each visitor, then delivers a chain for the matching platform without user interaction. The chains steal browser cookies, and in some cases Wi-Fi passwords or full device data. Sekoia says all of them use known, patched flaws, none are zero-days, and the iOS full-data chain is the DarkSword exploit kit. Compromised sites include a Ukrainian gas production company and a US chemical industry trade association.
- Exposure: Unpatched iPhones running iOS 16.0 to 17.4 or iOS 18.5, and Android devices running Chrome 139 through 145 (except 143) with vulnerable Adreno graphics drivers, are exposed when users visit the affected sites.
- Response: Update iOS and Chrome to current vendor releases, and apply the Android security patches that fix CVE-2025-27038. Check proxy logs for traffic to the domains Sekoia lists, including
static.flux-telemetry[.]comandad.starx-net[.]com, and reset sessions for users who visited affected sites. - Detections: Sekoia published domains, IP addresses, and file hashes, and says more IOCs and samples are available to CERTs and law enforcement on request.
OpenSSH "sshhole" flaw lets attackers log in without a password on Alpine
An attacker who can reach SSH on an affected Alpine Linux host can log in to accounts that have no password set, in a single connection, Alpine reports. The flaw, CVE-2026-123158, is in sshd built without PAM, which is Alpine's default openssh-server package: musl's crypt() returns * as an error for passwords over 256 bytes, and that matches the * stored for key-only accounts. It works when password authentication is on (the OpenSSH default), an account has * as its password field, and at least one other account has a $5$ or $6$ hash.
Every Alpine release from 3.16 through 3.24, plus edge, is affected. Official tiny-cloud images from 3.18.11 to 3.24.3 can be taken to root through their alpine user and passwordless doas. A public exploit exists. Zero Hunt reports exploitation in the wild, which could not be confirmed, and the flaw is not on CISA's KEV catalog.
- Exposure: Alpine hosts, and other musl-based systems without PAM, running sshd with password logins allowed. Containers that don't run sshd, and cloud-init images (which turn password logins off), are not affected.
- Response: Run
apk upgrade -U -a(fixed in Alpine 3.21.10, 3.22.8, 3.23.8, and 3.24.4, and in OpenSSH 10.6p2), then restart sshd: the upgrade doesn't restart the running daemon. If you can't update yet, setPasswordAuthentication no. - Detections: Alpine advises searching
/var/log/messagesfor sshdAccepted password forlines naming an account that has no password, such asalpineorroot, and treating the host as compromised if a login can't be ruled out. - Also reported by: Zero Hunt
Linux kernel IPv4 flaw allows remote code execution, Fortinet investigating
FortiGuard Labs reports that CVE-2026-98432, a use-after-free flaw in the Linux kernel's IPv4 networking code, could let an unauthenticated remote attacker run code with kernel privileges. Fortinet's advisory rates it critical (CVSS v3.1 9.8) and says it is not known to be exploited. The CVE is not on CISA's Known Exploited Vulnerabilities (KEV) catalog.
Attackers could trigger the flaw by sending crafted IPv4 TCP packets that carry IP options, though exploitation requires precise timing. The bug is a race over socket-option memory that is freed too early while a reply to a connection request is still being built. Fortinet says it is still investigating which of its products are affected, and lists dozens, including FortiOS, FortiGate Cloud, FortiMail, FortiManager, FortiAnalyzer, FortiSIEM, and FortiWeb.
- Exposure: Linux systems on kernels since 4.4 that don't yet have the fix, and the Fortinet products under investigation, which Fortinet has not yet confirmed as vulnerable.
- Response: Install your distribution's kernel security update and reboot so the patched kernel runs; Ubuntu 26.10 already includes the fix, and fixes for Ubuntu 22.04 and newer are in its proposed repository. Until then, limit inbound TCP from untrusted networks or apply Red Hat's interim nftables rule that filters IPv4 packets with IP options.
- Detections: FortiGuard published an IPS signature, Virtual Patch signature 10010015, and endpoint vulnerability signatures for CVE-2026-98432.
- Also reported by: SANS Internet Storm Center
More This Week
- Google Chrome update fixes multiple flaws, including code execution: Google's latest desktop update fixes multiple vulnerabilities, the Center for Internet Security reports. The most severe could let a malicious web page run code as the logged-on user; none is reported exploited. Update to 157.0.8103.38 or later (157.0.8103.38/.39 on Windows and Mac). CIS
- Gorbag exploits WinRAR flaw against Ukrainian justice and local government (CVE-2025-8088): ESET details Gorbag, a Russia-aligned group targeting Ukrainian justice, law enforcement, local administrations near the front line, and defense since February 2025. Since early 2026 it has used CVE-2025-8088, a WinRAR path traversal flaw fixed in WinRAR 7.13, to deliver a stealer that collects browser credentials, VPN configurations, and documents. ESET
- Scans continue for old Cisco IOS flaw (CVE-2023-20198): The SANS Internet Storm Center saw a honeypot probe, likely for CVE-2023-20198, running a "uname -a" command through the web interface. The flaw has been on KEV since October 2023. SANS Internet Storm Center
- WordPress Core (CVE-2026-119977): CISA added this time-of-check time-of-use race condition flaw to its Known Exploited Vulnerabilities catalog on Nov 3; federal agencies must apply the fix by Nov 6. CISA
- SonicWall SonicOS SSLVPN (CVE-2026-19868): CISA added this use of single-factor authentication flaw to its Known Exploited Vulnerabilities catalog on Nov 6; federal agencies must apply the fix by Nov 20. CISA
Updates on Earlier Stories
Refrigeration controller attacks spread from France to at least 10 more countries
Unit 42 reports that unidentified attackers gained access to refrigeration controllers across Europe and set them to defrost mode. The first wave hit roughly 700 French supermarkets overnight on Oct. 11-12, 2026. A second wave on Nov. 1-2 reached at least 10 more European countries and 100 to 115 retail locations. First covered in 26.43; 26.44 added Forescout's count of exposed controllers.
Attackers raised unit temperatures as high as 50°C (122°F), causing some fires and forcing stores to discard spoiled food. French authorities also disclosed that refrigeration systems at medical facilities were affected, with no evidence that blood products spoiled. The initial access vector is unconfirmed, though Unit 42 notes speculation about vulnerable software versions or default manufacturer credentials. The pro-Russian groups NoName057(16) and Z-Pentest claimed responsibility on Nov. 3, but Unit 42 says the claims are unverified.
- Exposure: Organizations in retail, food, and healthcare cold chains that run internet-exposed refrigeration controllers from the manufacturers named in the CERT-FR advisory; Unit 42 counted over 5,000 such exposed devices on Nov. 5-6, though exposure alone does not mean a device is vulnerable.
- Response: If you operate these controllers, take them off the public internet, replace any manufacturer default credentials, and apply vendor firmware fixes. CISA's Nov. 5 advisory warns of authentication bypass, arbitrary code injection, and weak configurations, so review it for hardening steps.
Credential Theft & Phishing
From the editor: device code phishing outlives the EvilTokens takedown
Six weeks after Microsoft's Sept. 22 takedown of EvilTokens, a detection built for its delivery chain started firing again, the editor writes. The emails still open with a TikTok redirect carrying the same unusual lzdlang=enFSmPWg parameter, pass through a Google DoubleClick hop, and end at Microsoft's real device code sign-in page, where the victim's own sign-in and MFA hand the attacker valid tokens. Six backend domains behind the lure pages appear to share one session counter, and their API matches EvilTokens path for path, which points to a shared platform serving several customers. Two of the backend domains were registered well before the takedown, so the platform may predate it.
- Exposure: Microsoft 365 tenants that allow device code sign-ins, which most users never need.
- Response: Check who actually uses device code flow in Entra sign-in logs, then block it with a Conditional Access policy (the Authentication flows condition) and exclude what legitimately needs it. If a user signed in through one of these lures, revoke their sessions and refresh tokens; a password reset alone doesn't end the attacker's access.
- Detections: The post includes Sentinel and Defender XDR queries, among them one for delivered messages carrying the redirect parameter, plus backend domains, redirector templates, and API paths.
Quick Hits
- Microsoft email threat landscape for Q3 2026: Phishing threats reached about 7.3 billion, with September up 41% from August. QR code phishing stayed far below its early-year peak, CAPTCHA-gated phishing rebounded on SVG attachments, and exceptions let spoofed malicious mail through. Microsoft Threat Intelligence
- Okta warns AI will accelerate credential stuffing attacks: Okta says AI coding agents now help attackers write configuration files for targeted services, likely increasing attack frequency. Its fix is passkeys, which stop stolen passwords from being replayed. Okta
- Fake business order emails lead to XLoader in-memory injection: Morphisec traces an order-themed business email lure to XLoader, which is injected into memory. Cryptika says the malware steals passwords and evades detection. Morphisec · Cryptika
Ransomware
Anubis ransomware attackers deploy new Lindnor backdoor, security killers
Symantec reports that attackers using Anubis ransomware hit a U.S. organization in October 2026 with a previously undocumented custom backdoor named Lindnor. Within about an hour, they dropped at least 13 executables built to terminate security services, plus five known vulnerable drivers. The encryption payload was blocked, and Symantec found no evidence that files were encrypted.
Lindnor ran as a service on two Microsoft SQL Server machines, disguised as a Chrome component (healthcontroller64.exe) and a VMware file (svchelper.exe). The attackers also installed AnyDesk, a legitimate remote access tool, and ran safemode.exe to register it to run in Safe Mode with Networking, when security software generally does not fully operate. Symantec did not determine initial access, and the backdoor also appeared on two other unrelated networks with no ransomware activity.
- Exposure: Windows environments; Symantec found the backdoor on Microsoft SQL Server hosts, and 52% of Anubis's roughly 96 claimed victims in the past 12 months are in the U.S.
- Response: If you find unexpected AnyDesk installs, new local accounts, or AnyDesk registered under the
SafeBoot\Networkregistry key, treat the host as compromised and investigate it. Check SQL Server hosts for services named like Chrome or VMware components that run from unusual folders. - Detections: Symantec published file hashes for Lindnor, the Anubis payload, the security killers, and the drivers, plus two command-and-control domains.
Ransomware Tracker
Ransomware and data-extortion groups: group · sectors when named · what's new this week
- 3AM · Education · Bitdefender describes a hybrid attack on European schools by the 3AM ransomware group, combining human operators with an agentic AI loop. Bitdefender
Threat Activity & Tradecraft
ARTEX agentic AI framework drives espionage and cloud-secret theft
CrowdStrike reports that VAPOR PANDA, a China-nexus actor, used the ARTEX agentic AI framework with DeepSeek and a second commercial large language model (LLM). Starting October 17, 2026, a single controller targeted more than 200 organizations, including foreign ministries, telecommunications providers, and U.S. pharmaceutical, semiconductor, and defense firms, and CrowdStrike confirmed post-exploitation activity at nine. A human operator steered parallel intrusions that used unsigned SAML assertions, cloud metadata theft, web shells, LAPS password reads, and forged mail tokens. CrowdStrike calls this separate from the Storm-0940 ARTEX activity covered in 26.44.
Separately, Microsoft Threat Intelligence reports that threat actors are using agentic tooling, including ARTEX, to search public GitHub repositories for exposed credentials, application secrets, and AI API keys. In one case, an agent went from scanning GitHub to exfiltrating Azure Key Vault data in five minutes. From October 28, 2026, Microsoft saw successful service principal sign-ins from 199.19.77[.]84 across multiple tenants, followed by access to Key Vaults and Storage Accounts and scripts run on Azure virtual machines through Run Command.
- Exposure: Applications that accept unsigned SAML assertions, cloud instances that allow IMDSv1, and hosts without endpoint detection and response (EDR) coverage; and organizations with secrets in public GitHub repositories, especially Entra ID service principals with access to Key Vault or Storage.
- Response: Enforce SAML signature validation and require IMDSv2. Treat any secret that appeared in a public repository as compromised: rotate it, then rotate everything in the Key Vaults that application could read. Review Entra service principal sign-ins for 199.19.77[.]84, and hunt for
artxorartexstrings in command lines,artx-prefixed cloud roles, and theC:\ProgramData\artx5v\directory. - Detections: CrowdStrike published IOCs for the BLOODALCHEMY backdoor, and Microsoft published advanced hunting queries, Defender detection coverage, and indicators for the October 28 sign-ins.
- Also reported by: Google Threat Intelligence Group, Microsoft Threat Intelligence, SOCRadar
LLMJacking attacks hijack cloud AI access for resale
Google Threat Intelligence Group (GTIG) reports that Mandiant responded to multiple 2026 intrusions in which attackers hijacked enterprise access to large language models (LLMs), a tactic called LLMJacking. Victims in the financial services, retail, technology, and transportation sectors saw fraudulent charges above $100,000 in a single day, and cumulative charges at some organizations reached $1 million before access was revoked.
The dominant entry point was exposed or stolen cloud identity tokens, including IAM credentials, service account keys, and GitHub personal access tokens. Attackers then enabled LLM services, created extra keys, and added backdoor accounts. Some resold the access through "LLM Mixers," gateways that pool stolen credentials. In one case, a follow-on cluster abused CVE-2026-40217 (not on KEV) in a LiteLLM gateway to escalate standard keys to administrator.
- Exposure: Organizations with cloud LLM services (including Azure
Microsoft.CognitiveServices, AWS Bedrock, and Google Gemini APIs), exposed model endpoints, or LiteLLM gateways, especially where keys sit in public repositories or LLM services were never enabled. - Response: Alert on any unauthorized enablement of LLM services, new model deployments, or changes to diagnostic logging, and set API quotas and spending caps. For LiteLLM, check for and apply a release that fixes CVE-2026-40217.
- Detections: GTIG published Azure, AWS, and GCP audit log hunting guidance, Google SecOps rule names, and IOCs in a GTI Collection.
- Also reported by: CyberVeille
BITTER adds PapRAT backdoor to espionage campaign
Fortinet's FortiGuard Labs reports a campaign against South Asian organizations that uses PowerShell scripts and a custom remote access trojan (RAT) it calls PapRAT. Fortinet attributes the activity to BITTER, a suspected South Asian cyberespionage group active since at least 2013, with high confidence. BITTER has a history of targeting government, energy, and engineering sectors.
The campaign uses three infection chains: a LNK file that abuses the Windows ftp.exe utility and scheduled tasks, a VHDX disk image that silently launches an MSI installer, and a URL shortcut that pulls a LNK file over WebDAV. The third chain also delivers WmRAT, a second implant. PapRAT has C++ and Rust versions, and the Rust version raises its command set from eight to eleven, adding PowerShell execution, sleep, and exit. Commands include remote shell, file upload and download, and screenshots.
- Exposure: Windows systems at government, energy, and engineering organizations; Fortinet's telemetry shows targets concentrated in South Asia.
- Response: Where these tools are not needed, block or alert on LNK files, URL shortcuts, VHDX images, and MSI packages arriving through unsolicited messages. Also hunt for
ftp.exelaunching shell commands and for scheduled tasks that runerr.logortalk.logthroughcmd. - Detections: Fortinet published IOCs (hostnames, URLs, and file hashes) and lists FortiGuard antivirus signatures for the LNK files, PapRAT, and WmRAT.
- Also reported by: CyberVeille
Worth Watching
- Suspected MuddyWater uses Israel Tax Authority lures and SimpleHelp: Cato Networks reports suspected Iranian MuddyWater activity using Israel Tax Authority lures and the SimpleHelp remote tool. Separately, CrowdStrike found Iran-nexus STATIC KITTEN holding a modified FortiOS exploit that creates backdoor admin accounts. Cato Networks · CrowdStrike
- North Korean attackers move $6.34 billion out of Binance after compromising developer laptops: About $6.34 billion left Binance hot wallets on October 29 after North Korean attackers compromised developer laptops through a tampered Homebrew install and a fake job interview; partners froze or intercepted about $2.85 billion. The FBI attributes the attack to North Korea; Binance covers user losses. SafeDep · ThreatMon · OpenSourceMalware
- Fake-download campaign uses trojanized installers to deliver monetization payloads: Microsoft says trojanized installers posing as cracked software install proxyware, distributed-computing, and adware-style payloads, often without informed consent. Cyber Press describes the same campaign as secretly monetizing victims' internet connections. Microsoft Threat Intelligence · Cyber Press
Further Reading
- Microsoft analyzes what exploited vulnerabilities have in common: Only about 2,000 of roughly 383,000 disclosed CVEs (about 0.5%) have been exploited. Takeover classes (remote code execution, privilege escalation, authentication bypass) account for 78% of them, so Microsoft advises prioritizing those on exposed servers. Microsoft Threat Intelligence
- Why reputation feeds miss software supply chain attacks: OpenSourceMalware found 62.2% of 450 day-zero indicators had zero VirusTotal detections. Across ten 2026 campaigns, 71.3% of indicators were still blind spots months later. OpenSourceMalware
Every item links to its original source. Spot an error? Reply or leave a comment. Know a defender who'd find this useful? Forward it along.
GPP Labs · Threat Intelligence & Detection Engineering
GPP // Genuine People Personality™