Security news, decoded.

Ivanti has released an urgent patch for a high-severity authentication bypass flaw in its Connect Secure (ICS) remote access gateway, warning that attackers are already exploiting it. The vulnerability, cataloged as CVE-2026-104935 and discovered by Volexity, affects ICS versions 22.8R2.4 and earlier, 25.1.2.2 and earlier, and 25.1.1.1 and earlier. It allows unauthenticated remote attackers to run arbitrary commands on vulnerable appliances without any user interaction, making it an easy entry point into corporate networks. Ivanti confirmed limited real-world exploitation and is advising customers to upgrade immediately and routinely run its Integrity Checker Tool, while Shadowserver reports more than 5,500 ICS instances still reachable from the internet. This marks yet another in a long string of Connect Secure zero-days abused by attackers in recent years, following flaws like CVE-2023-46805, CVE-2024-21887, CVE-2024-21893, CVE-2025-0282 and CVE-2025-22457.

Docker has patched a critical vulnerability, tracked as CVE-2026-105445, in Docker Hub that allowed attackers to hijack another user's personal access token (PAT) by converting their own account into an organization and designating the victim as its owner. Because the reassignment process didn't rotate the token's secret or require the target's consent, the attacker could then authenticate as that user within the PAT's scope and push, pull, or delete images in their repositories and connected organizations — a serious supply-chain risk since it could enable publishing malicious images under a trusted identity. The bug, present since August 26, 2024, was reported on September 30, 2026, and fixed in production within four hours; Docker says a review of 304,000 token-based pushes found no sign of abuse. Affected accounts were notified on October 2 and had their transferred tokens deactivated on October 8, while Docker Hub now requires explicit consent before naming someone an organization owner and no longer auto-transfers PATs during conversions. Docker clarified the flaw did not grant Docker Hub web console access or let attackers alter passwords, email addresses, or two-factor authentication settings, so account takeover in the traditional sense wasn't possible. Users who never received a notification from Docker are considered unaffected and don't need to act.

Check Point Research examined LiteLLM, a popular open-source gateway for routing requests to LLM providers, and found serious flaws in version 1.77.7, now fixed in 1.83.9. The core issue, tracked as CVE-2026-35029, was a missing authorization check on the /config/update endpoint, letting any logged-in user—even one with minimal privileges—rewrite the proxy's live configuration. From there, attackers could inject environment variables to read arbitrary files, overwrite admin login credentials to seize full control of the gateway, or register a malicious Python module from cloud storage to achieve remote code execution. A second flaw, CVE-2026-59823, resided in the "Clientside LLM Credentials" feature and let attackers abuse the userconfig parameter to redirect outgoing API calls via SSRF, capturing the provider's real API keys on an attacker-run server. Researchers also flagged weakly hashed, unsalted passwords and exposed password hashes as additional risks, underscoring that even fast-moving AI infrastructure projects still need solid access-control and input-validation fundamentals.

SAP's October 2026 security update rolls out fixes for 20 vulnerabilities, headlined by a maximum-severity flaw in the Cloud SDK for Python used to build AI agents and apps on SAP's Business Technology Platform. Tracked as CVE-2026-89111, the bug allows an unauthenticated remote attacker who controls the tenantsubdomain parameter to redirect OAuth2 token requests to a server they control, exfiltrating client credentials from multi-tenant BTP applications. Organizations unable to apply the patch immediately are advised to restrict tenantsubdomain values to a known allowlist as a temporary workaround. The same update batch also closes a critical access-control bug in SAP NetWeaver Visual Composer (CVE-2026-89110) and a critical privilege-escalation issue in SAP Manufacturing Integration and Intelligence (CVE-2026-89108), continuing a string of severe SAP flaws patched in recent months, including CVE-2026-58231 in Commerce Cloud and the OVERPASS kernel bug CVE-2026-44756.

Bishop Fox researchers disclosed CVE-2026-58113, a reflected cross-site scripting flaw in the /auth/ login-redirect page of Siemens Teamcenter's Active Workspace interface. Because the page reflects query parameters unencoded into an HTML form and then auto-submits that form on load, an attacker-crafted link executes injected JavaScript the instant a logged-in victim opens it, no click or interaction beyond opening the URL required. During an authorized assessment, the researchers chained this into a state-changing call to Teamcenter's workflow API, silently granting one account surrogate-delegate rights over another user's session, which in a PLM system could translate into unauthorized approval of engineering changes or release of design data. Siemens rates the issue 8.5 (High) under CVSS 4.0 and has released fixed builds across the V2412, V2506, V2512, and V2606 branches, documented in advisory SSA-157465; no in-the-wild exploitation has been reported, but organizations are urged to update promptly and audit existing surrogate delegations for unexpected entries.

GitLab has released emergency fixes for a maximum-severity flaw, CVE-2026-97229, found in the regular expression compiler used by its CI/CD pipeline configuration. The bug, an integer overflow rated CVSS 10.0, could let an unauthenticated attacker run arbitrary code on a self-managed GitLab server simply by submitting a malicious regex inside a CI/CD config file. It impacts both GitLab Community Edition and Enterprise Edition across the 19.2, 19.3 and 19.4 release lines, and GitLab is telling every self-hosted admin to upgrade immediately to 19.4.2, 19.3.5 or 19.2.8, published on October 12, 2026. GitLab.com already runs patched code and Dedicated customers need not act, while the issue was disclosed through HackerOne by researcher joaxcar, who also flagged a separate medium-severity open-redirect bug (CVE-2026-7767) in GitLab's Kubernetes proxy.

On the night of October 11, attackers remotely took over refrigeration controllers made by Danfoss at hundreds of supermarkets across France, pushing fridge and freezer temperatures as high as 50°C and forcing chains to discard spoiled food. The intrusions hit Danfoss's AK-SM and AK-SC controller lines, which manage cooling systems and can be administered remotely by maintenance firms; Les Mousquetaires (Intermarché, Netto), Coopérative U and Carrefour all reported affected stores, with Les Mousquetaires estimating losses in the tens of millions of euros. France's CERT-FR has urged operators to disconnect these controllers from the internet or tightly restrict access, noting that many compromised units were older, now-unsupported models like the AK-SM 800 and AK-SC 355, while some current AK-SM 800A devices running outdated firmware were also affected. Neither Danfoss nor French authorities have disclosed the exact vulnerability or attacker identity, though a government source mentioned ANSSI identified a flaw in the vendor's software and separate reports suggest similar incidents in Spain, Portugal, Poland, Germany and Belgium; a Paris prosecutorial inquiry is now underway.

Progress Software has disclosed 31 security flaws in MarkLogic Server, including five rated critical, with fixes already shipped in MarkLogic 11.3.7 and 12.1.0 since September 10. Two of the critical bugs stand out: CVE-2026-85947 and CVE-2026-85948, both scored 9.9, which could let a user with only limited delegated management rights (such as forest-management or manage-user privileges) create a persistent administrator account and effectively seize full control of the server. Other high-severity issues (CVE-2026-85955, CVE-2026-85956, CVE-2026-9191) allow elevated code execution or unauthorized cross-database access, while additional flaws affect SAML, OAuth, Kerberos, cloud credential exposure, and logging of sensitive data. Progress has found no evidence of in-the-wild exploitation and none of the CVEs appear in CISA's Known Exploited Vulnerabilities catalog, but customers running 11.x before 11.3.7 or 12.x before 12.1.0 — including those who already applied August's patches — are urged to upgrade promptly, since no workaround exists.

GitLab has shipped fixes for a maximum-severity vulnerability in its Community Edition and Enterprise Edition products, tracked as CVE-2026-97229 and rated CVSS 10.0. The bug stems from an integer overflow tied to inadequate validation of regular expressions in CI/CD configuration, which under certain conditions could let an unauthenticated attacker execute arbitrary code on the GitLab server. It impacts versions 19.2 before 19.2.8, 19.3 before 19.3.5, and 19.4 before 19.4.2, with fixes delivered in 19.2.8, 19.3.5, and 19.4.2. The same update also resolves three high-severity issues: CVE-2026-92883 (CVSS 8.7), a stored XSS flaw letting an authenticated user run JavaScript in other users' sessions via a crafted repository; CVE-2026-16301 (CVSS 8.5), an authorization bypass letting a Developer-role user trigger CI/CD pipelines on protected branches; and CVE-2026-14814 (CVSS 8.2), a loose regex check allowing cache poisoning of raw repository file responses that could expose protected environment variables through manipulated CI/CD pipeline runs. There is currently no evidence of exploitation for the new flaws, though GitLab has faced several other critical patches recently, including a CVSS 9.9 AI Gateway sandbox-escape bug and a CVSS 10.0 file-read issue from last month that attackers began probing within hours of disclosure.

Attackers are already probing a perfect-10 CVSS vulnerability in SonicWall's SMA 1000 Secure Mobile Access gateway, despite the vendor's claim of no confirmed in-the-wild exploitation. The flaw, CVE-2026-102255, is a server-side request forgery bug in the appliance's WorkPlace remote-access portal, caused by an unexpected alternate access route that lets an unauthenticated attacker force the device to make requests on their behalf and reach internal systems. Honeypot operators Previdian and Defused reported seeing unauthenticated SSRF probing against SonicWall SMA decoys within days of SonicWall's October 6 advisory, and researchers at SOCRadar warn the device's internet-facing position could let intruders pivot into normally unreachable internal services. SonicWall has published three other CVEs alongside this one and released a hotfix, urging SMA 1000 customers to patch immediately, while noting that SMA 100 devices and firewall-based SSL-VPNs are not impacted.

A cyberattack hit refrigeration systems at several major French supermarket chains overnight, remotely forcing cooling units to defrost and causing widespread spoilage of fresh food. According to French media, hackers compromised Danfoss refrigeration controllers running outdated software, exploiting their built-in automatic defrost function to raise temperatures across an estimated 250 to 350 stores belonging to Intermarché, Coopérative U, Carrefour Market, Monoprix, Leclerc, and Super U. Coopérative U's president publicly confirmed the incident and called the targeting of food-preservation infrastructure unprecedented and unacceptable, while the Paris prosecutor's office has opened a probe to identify victims and assess the resulting losses and reported fires. It remains unclear how the attackers gained access, though CISA flagged a high-severity authentication bypass flaw, CVE-2025-41450, in Danfoss's AK-SM 8xxA Series hardware back in May 2025; separately, retailer Ahold Delhaize also disclosed an unrelated cybersecurity incident affecting its own systems around the same time.

This week's roundup opens with active exploitation of a critical Atlassian flaw, CVE-2026-21589 (CVSS 9.3), which lets attackers reach sensitive files across Bitbucket, Confluence, Jira Software/Service Management, Bamboo, Crowd, Crucible, and Fisheye Data Center products; attacks reportedly started just two hours after watchTowr published technical details. Other headlines include the FBI's third arrest tied to the ShinyHunters extortion crew, Google's discovery that attackers hijacked Ghana, American Samoa, and Sierra Leone ccTLDs to forge HTTPS certificates for Google and other major brands, and a flaw in the GoBalance load balancer used to hijack Tor .onion sites, including the Dread darknet forum. Also covered: an updated P7 DarkSword iOS exploit kit stealing crypto wallets and keychain data, malware families embedding text designed to trick AI-based analysis tools, the FBI's takedown of hacking infrastructure tied to China's Flax Typhoon, and a long list of urgent CVEs across Dell, Chrome, SonicWall, Elasticsearch, Apache Struts, Veeam, Cisco, Splunk, and WordPress plugins that administrators should patch without delay.

Check Point Research's latest weekly bulletin covers several major breaches and newly patched flaws. Denmark's population registry, UK retailer ASOS, Ukrainian grocery chain ATB, and Osaka Metropolitan University all disclosed intrusions exposing millions of personal records, with groups like Xuanye and DataSuckers claiming responsibility. The report also details AI-assisted attack campaigns, including the ARTEX toolkit used against South Korean banks and the CyberXero initial-access broker hitting Ukrainian infrastructure. On the patching front, Veeam fixed a critical RCE in Backup & Replication 12 (CVE-2025-64393, CVSS 9.4), Citrix confirmed active exploitation of a NetScaler ADC/Gateway flaw (CVE-2026-88779, CVSS 8.7), Cisco addressed a critical NX-OS access-control bug (CVE-2026-76455, CVSS 9.8), and Atlassian released patches for a critical file-read vulnerability affecting eight Data Center products (CVE-2026-21589, CVSS 9.3). Separately, US authorities warned that the FortiBleed credential-theft campaign has compromised over 86,000 Fortinet devices worldwide, with access being resold to ransomware affiliates.

Patchstack has detected active exploitation of a critical flaw in Bricksforge, a commercial add-on used by around 6,000 WordPress sites running the Bricks builder. Tracked as CVE-2026-85097 with a CVSS score of 9.8, the bug lets unauthenticated attackers upload arbitrary files and run malicious code, potentially leading to complete compromise of affected websites. The issue affects versions up to and including 3.1.8.9 and was fixed in version 3.1.8.10. Patchstack says it began observing in-the-wild attack attempts on October 7, with telemetry showing two distinct clusters of automated requests, one involving dozens of IPs firing identical payloads and another testing multiple file-extension and encoding variations, suggesting coordinated but not necessarily single-actor automated scanning.

GitLab has issued an urgent patch for a maximum-severity flaw, CVE-2026-97229, found in its CI/CD Regular Expression Compiler, which handles parsing and evaluation of regex patterns inside pipeline configuration files. The bug is rooted in an integer overflow that could let an unauthenticated attacker run arbitrary code on a GitLab server simply by submitting a maliciously crafted regular expression in a CI/CD config, due to inadequate validation of these expressions. The issue affects self-managed GitLab Community Edition and Enterprise Edition releases from 19.2 up to 19.2.8, 19.3 up to 19.3.4, and 19.4 up to 19.4.2, and the company is urging all self-hosted customers to upgrade immediately, noting that GitLab.com and GitLab Dedicated are unaffected or already protected. The flaw was reported through GitLab's HackerOne bounty program by researcher 'joaxcar' and follows several other severe GitLab vulnerabilities fixed in recent months, including a 2FA bypass and a path-traversal bug that CISA confirmed was actively exploited.

Attackers have begun probing a maximum-severity SonicWall Secure Mobile Access (SMA) 1000 vulnerability just days after it was publicly disclosed and patched. The flaw, CVE-2026-102255 (CVSS 10.0), is a pre-authentication SSRF bug that lets an unauthenticated remote attacker force the appliance to make requests on their behalf, reaching internal services and performing unauthorized actions. SonicWall shipped fixes on October 6, 2026 via hotfix builds 12.4.3-03670 and 12.5.0-03082, but threat-intel firms Defused Cyber and Previdian spotted exploitation attempts starting October 8, with Previdian logging 135 attempts from two US-based IP addresses; SonicWall's advisory has not yet been updated to confirm active exploitation. This follows an earlier wave of zero-day attacks in June 2026 against the same product line, tracked as CVE-2026-15409 and CVE-2026-15410, which a group called UTA0533 used to deploy web shells and the KNUCKLEBALL malware, with Rapid7 reporting that compromised appliances were used to steal credentials, session data, and MFA seeds before pivoting into internal networks.

Patchstack reports active in-the-wild exploitation of a maximum-severity flaw in the Bricksforge plugin for WordPress, tracked as CVE-2026-85097 (CVSS 10.0). The bug lets unauthenticated attackers abuse the 'temporaryFileUploads' parameter — bypassing MIME checks with a GIF/PHP polyglot file and mismatched file paths — to upload and run arbitrary PHP code, giving them full remote code execution on affected sites. All Bricksforge versions through 3.1.8.9 are vulnerable; the vendor shipped fixes in 3.1.8.10, with 4.0.1 recommended as the current safe release. Patchstack first spotted attack traffic on October 7, 2026, from dozens of rotating IPs testing various PHP extension-evasion tricks, and separately, attackers are also exploiting XSS bugs in WPC Product Bundles for WooCommerce (CVE-2026-93836) and Ninja Forms (CVE-2026-94504) to plant hidden admin accounts and backdoors on WordPress sites.

Attackers have begun targeting a maximum-severity flaw in SonicWall SMA1000 appliances just days after the vendor released a fix. The bug, CVE-2026-102255 (CVSS 10), is a pre-authentication server-side request forgery issue in the SMA1000 Work Place interface that lets an unauthenticated remote attacker trick the device into making requests on its behalf, exposing internal functionality to unauthorized use. SonicWall disclosed the flaw on October 6 and told customers running the 6210, 7210, and 8200v models to upgrade to firmware 12.5.0-03082 or 12.4.3-03670. Threat intel firms Previdian and Defused reported seeing the first exploitation attempts against honeypot sensors as early as October 8, while Horizon3.ai cautioned that successful attacks could lead to full device takeover.

{"title":"CyberXero Uses Claude Code AI Agents to Steal 628,000+ Ukrainian Records","seotitle":"CyberXero exploits WordPress via Claude AI","seodescription":"Active since July 2026: CyberXero used Claude Code and PentAGI with Cobalt Strike to exploit CVE-2026-63030, CVE-2026-60137, CVE-2026-4815, stealing 628K+ Ukrainian records."}

US and Jordanian authorities each detained a suspected member of the ShinyHunters hacking collective last week. The FBI arrested Canadian cybersecurity professional Edward Dubrovsky in Pennsylvania, whom Director Kash Patel linked to last month's intrusion into the FBI's own job application portal. Separately, Jordan detained Saif al-Din Khader, allegedly the ShinyHunters figure known as "Rey," who is reportedly now cooperating with the FBI to identify other group members. The arrests come as fallout from the FBI breach continues, including Accenture losing its FBI contract after failing to patch the Oracle PeopleSoft systems that ShinyHunters exploited to steal data on nearly all FBI special agents.

This week's roundup highlights that the FortiBleed campaign against Fortinet FortiGate firewalls and SSL VPN gateways remains active, with a joint FBI and U.S. Secret Service advisory warning that some victim organizations have been locked out of their own appliances; SOCRadar has confirmed over 86,644 compromised devices across 194 countries. Other notable items include active exploitation attempts against a critical Atlassian Data Center flaw (CVE-2026-21589), CISA adding an exploited NetScaler denial-of-service bug (CVE-2026-88779) to its known-exploited list, Microsoft's emergency Exchange Server patch for a mailbox-access vulnerability (CVE-2026-96940), and SonicWall fixing a pre-auth SSRF issue in SMA 1000 appliances (CVE-2026-102255). The digest also previews the October 2026 Patch Tuesday forecast following last month's record 973 CVEs from Microsoft, alongside breach news from Denmark's population registry and ASOS, plus a Dell System Update root-privilege flaw (CVE-2026-86360).

watchTowr Labs dissects the messy patch saga that followed PaperCut's 27 August advisory about in-the-wild exploitation of version 26.0.3. Starting from bare IOCs with no CVE or patch, researchers independently rediscovered the exploited flaws, later confirmed as CVE-2026-82078 (post-auth RCE) and CVE-2026-81578 (auth bypass), then found that each successive PaperCut patch introduced a new bypass: WT-2026-0141, WT-2026-0142, and finally WT-2026-0143, an authentication bypass abusing Tapestry's Setup Wizard pages via the Home page that still worked against build 26.0.4-PO build 76508. Combined with a freshly discovered post-auth RCE in the Scan-to-Fax feature (WT-2026-0144, CVE-2026-82077, a path traversal in the fax connector binary path), the pair forms a complete unauthenticated remote code execution chain against that build. PaperCut has since shipped build 76530 and version 26.0.5 to close these gaps, and watchTowr released a Detection Artefact Generator to help admins check exposure.

Attackers are actively abusing two unpatched flaws in the AhsayCBS backup management platform, used widely by MSPs and system integrators, to drop webshells and cryptomining malware. Huntress researchers detected the campaign on October 7, affecting at least five organizations, with intruders first exploiting the authentication-bypass bug CVE-2026-105133 (which already has a public exploit) and then chaining it with the OS command-injection flaw CVE-2026-105134 to gain code execution. Although the vendor listed both issues as resolved in version 10.3.2, Huntress confirmed that the current release, 10.3.4, remains vulnerable as well. Once inside, attackers installed JSP webshells and an XMRig miner disguised as edge.exe, maintained persistence through a fake Microsoft Edge update service, and used a likely AI-generated PowerShell script to hide the mining process whenever Task Manager was opened, while also deploying the known-vulnerable WinRing0x64.sys driver to boost mining performance. With no official fix yet available, Huntress urges admins to lock down the AhsayCBS management interface to trusted IPs, hunt for compromise indicators, and fully restore from clean backups if a breach is found, since attackers may have planted extra backdoors for long-term access.

Security researchers at iVerify have uncovered a new strain of the DarkSword iOS exploit kit, dubbed P7 DarkSword after the "p7" variable prefix found in its modified code. Compared to earlier versions first documented in March by GTIG, iVerify, and Lookout, this variant shrinks its on-device footprint, adds theft of keychain and cryptocurrency wallet data, and introduces two-way command-and-control communication through implants injected into the SpringBoard process. The kit, originally a commercial exploit chain targeting iOS 18.4 through 18.7, apparently leaked onto a secondary market and is now used by financially motivated actors and state-linked groups—including Turkish vendor PARS Defense and Russia's Star Blizzard (COLDRIVER)—against targets in Saudi Arabia, Turkey, Malaysia, and Ukraine. Censys further identified open directories tied to DarkSword and its companion wallet-stealing kit Coruna, revealing two previously unlinked flaws, CVE-2025-24201 (WebKit sandbox escape, fixed in iOS 18.3.2) and CVE-2025-31200 (Core Audio code execution, fixed in iOS 18.4.1), as well as a Chinese-speaking exploitation-as-a-service operation with victim data including 11 recovered crypto seed phrases.

Florida, Iowa, Montana and Nebraska filed consumer-protection lawsuits against TP-Link Systems on October 6, joining Texas in a legal campaign that now spans five states. The suits accuse the company of overstating the security of its products and exaggerating its separation from Chinese parent company TP-Link Technologies, while its Tether, Tapo, Deco and Kasa Smart apps allegedly gather data that could be exposed under Chinese intelligence law. As evidence, the complaints point to real-world attacks, including a Chinese-linked botnet built largely from compromised TP-Link routers and a Russian GRU campaign that exploited CVE-2023-50224 on end-of-life devices, plus end-of-life models like two Archer AX21 versions still marketed as secure. A day later, 21 state attorneys general sent the FCC a letter urging scrutiny of TP-Link's pending router approvals, while researchers at SEC Consult separately disclosed five critical flaws—CVE-2025-30237, CVE-2025-30238, CVE-2025-30239, CVE-2025-30240 and CVE-2025-30241—in roughly 65 ISP-supplied TP-Link models (HomeShield-enabled HB, HX, HC, EB, EC, EX, XC, XX and VX lines), for which fixes are rolling out through internet providers rather than direct downloads.

Security researchers from the V12 team have released a fully functional exploit, dubbed AnyPwn, for a pre-authentication remote code execution bug in AnyDesk Linux that lets an attacker gain root access before any connection is approved. The flaw is a heap buffer overflow in AnyDesk's session protocol, caused by unchecked 32-bit integer arithmetic that lets a crafted packet size wrap to zero, triggering a tiny allocation while a large length is still recorded, which attackers can abuse via a ROP chain to execute arbitrary commands as root. The researchers notified AnyDesk in June, and the vendor silently patched it in version 8.0.3 without assigning a CVE or publishing a security advisory, describing the fix only as a crash-related bug fix. The exploit, published on GitHub on October 8, specifically targets direct TCP connections on port 7070 in build 8.0.2 and is probabilistic, while AnyDesk states that macOS and Windows are unaffected and that relay-based connections are not exploitable, though the researchers showed the vulnerable code path is still reachable through relays. Administrators should update to AnyDesk Linux 8.1.0 or restrict access to port 7070 if an immediate update isn't possible.

Security researchers at Huntress report active exploitation of two AhsayCBS backup-software vulnerabilities, tracked as CVE-2026-105133 (an authentication bypass, CVSS 5.5) and CVE-2026-105134 (an OS command injection in the Replication Receiver component, CVSS 9.3). By combining both flaws, attackers can skip login checks and run arbitrary commands remotely. Huntress says intrusion attempts started on October 7, 2026, with at least five organizations hit by October 8, leading to reconnaissance, web shell deployment, and installation of XMRig cryptominers disguised under the filename edge.exe to mimic Microsoft Edge. The attackers also dropped an apparently AI-written PowerShell script that forces the mining process to pause whenever Task Manager is opened and kills Task Manager overnight if left running too long, and in one case used certutil.exe to fetch the vulnerable WinRing0x64.sys driver for kernel-level access. Although Ahsay's advisory claims the issues are fixed in version 10.3.4, Huntress found that release still exploitable, effectively making these zero-days; until a real patch ships, admins are urged to restrict access to the AhsayCBS management interface to trusted IPs or VPN connections.

A maximum-severity SSRF vulnerability in SonicWall SMA1000 appliances, CVE-2026-102255, is already being targeted by attackers just three days after a patch was issued. The bug sits in the Appliance WorkPlace interface of the SMA1000 6210, 7210 and 8200v models and lets an unauthenticated remote attacker force the device to make requests on their behalf, reaching internal services and performing unauthorized actions; SMA 100 series devices and firewall SSL-VPN are not affected. Security researcher Ryan Dewhurst of Previdian reported honeypot traffic matching this flaw, with crafted OPTIONS requests hitting the appliance's local CouchDB instance and trying to invoke an internal rewrite function using default admin:admin credentials, though it's unclear if any attempts actually succeeded. This follows a pattern of repeated SMA1000 zero-day abuse this year, including custom malware deployments and ransomware-linked intrusions tied to other CVEs, with Shadowserver counting over 400 SMA1000 devices currently exposed online.

Security researchers at Lava disclosed a high-severity bug in NVIDIA's DCGM Exporter, tracked as CVE-2026-47483 and scored 8.2 on CVSS, that allows attackers without any credentials to overwhelm and crash the GPU monitoring service by flooding it with concurrent requests to exposed Go profiling (/debug/pprof/) endpoints bundled with the official container. The researchers scanned the internet and found over 2,000 servers, covering more than 12,000 GPUs (including NVIDIA Blackwell Ultra B300, H200, H100, RTX 5090 and 4090 models) across nearly 300 organizations, publicly exposing metrics without authentication, with customers of cloud providers such as Voltage Park, Lambda, Northern Data and DigitalOcean affected. They also found that Prometheus Node Exporter, often deployed alongside DCGM Exporter, leaked detailed hardware, firmware, OS and network fabric information that could help attackers map out and target infrastructure. NVIDIA has fixed the issue in DCGM Exporter version 4.8.2, where profiling is now opt-in, and recommends disabling the --enable-pprof flag and restricting exporter access to private networks.

CISA has added five more vulnerabilities to its Known Exploited Vulnerabilities catalog after confirming they were abused by the Chinese state-linked group Flax Typhoon. The newly listed bugs span ProFTPD (CVE-2015-3306, CVSS 10.0), ONLYOFFICE Docs (CVE-2021-3199), Strapi (CVE-2023-22894), Apache Struts (CVE-2016-3081), and ISC BIND (CVE-2015-5477), ranging from arbitrary file access and remote code execution to sensitive-data exposure and denial-of-service. The listing lines up with a joint alert from the US, Australia, Canada, Japan, New Zealand, Spain, and the UK describing how the China-based firm Integrity Technology Group has leveraged eight flaws total—including these five plus three already-known CVEs tied to Shellshock, Ivanti Pulse Connect Secure, and GitLab—to break into networks, maintain access via VPN tools, and steal emails and credentials. Federal agencies in the US are required to remediate or retire the affected software by October 11, 2026, given confirmed active exploitation.

This week's security digest covers a wide range of incidents. South Korean President Lee Jae Myung indicated that artificial intelligence may have been involved in recent attacks against local banks that exposed customer data, with CrowdStrike linking the intrusions to a likely Chinese-speaking, financially motivated group that left behind Claude Code and ARTEX-related files. Separately, researchers at Black Lotus Labs uncovered PoeLLM, a cryptomining botnet that hides its command server address inside a GitHub-hosted poem, allowing operators to relocate infrastructure simply by editing the verses. GitGuardian reported that the GhostAction supply-chain campaign has now hit 772 additional GitHub repositories, stealing over 2,500 credentials, while Socket and Sonatype flagged a compromised Tensorlake npm SDK release carrying a Shai-Hulud-style credential-stealing worm. On the vulnerability side, CVE-2026-47483 was disclosed in Nvidia's DCGM Exporter, exposing telemetry from more than 12,000 GPUs across roughly 2,100 internet-facing hosts before being fixed in version 4.8.2. The roundup also notes legal outcomes, including a 40-year prison sentence for Empire Market co-founder Raheim Hamilton and a jury conviction against a hacker behind the Uranium Finance crypto thefts, plus CISA's revised cyber retention pay rules and a new OT security proposal for federal agencies.

Security firm Huntress reports that threat actors are actively abusing two unpatched security flaws in Ahsay Systems' AhsayCBS backup management console to achieve remote code execution. The bugs, CVE-2026-105133 and CVE-2026-105134, let attackers tamper with function parameters to sidestep authentication and inject operating-system commands; they were first flagged on October 4 alongside public exploit code, initially affecting all builds up to 10.3.2, though Huntress confirms the newest release, 10.3.4, is also vulnerable. The firm says intruders are combining both flaws—exploiting an authentication-bypass weakness in the Replication Receiver API tied to CVE-2026-105134—to drop JSP webshells, deploy disguised XMRig cryptominers, and set up persistence via a fake Microsoft Edge Update service, with at least five organizations hit by October 8. Since no fix exists yet, Ahsay customers are urged to lock down access to the AhsayCBS management interface, limiting it to trusted IPs or VPN connections, and to check for intrusion indicators.

Citrix is urging administrators to immediately update NetScaler ADC and NetScaler Gateway appliances after disclosing a new critical flaw tracked as CVE-2026-107406. The bug is caused by a memory overflow issue that could let attackers achieve remote code execution or crash the device through a denial-of-service condition. Devices are only at risk if they are set up as a SAML Identity Provider or Service Provider. Citrix says it has no evidence of active exploitation so far, but it recommends upgrading to NetScaler ADC/Gateway 14.1-73.46 or later, 13.1-64.29 or later, 14.1-FIPS 14.1-73.46 or later, and 13.1-FIPS/13.1-NDcPP 13.1.37.283 or later, noting that over 21,000 internet-facing NetScaler instances remain exposed and that similar flaws have been repeatedly targeted by attackers throughout the year.

At Pwn2Own Ireland 2026 in Cork, three separate research teams successfully remote-hacked a fully patched Google Pixel 10 on October 8, with Trend Micro's Zero Day Initiative (ZDI) confirming all three entries as working exploits. Xint earned $150,000, a group led by Dimitrios Valsamaras received $112,500 for a chain combining a known bug with a zero-day, and Ikotas Labs took the top prize of $300,000 for a multi-bug chain, helping it secure the overall "Master of Pwn" title. The exploits reportedly worked through the phone's default browser or via NFC, Wi-Fi, Bluetooth or baseband radios, though technical details remain undisclosed while Google and other vendors get a 90-day window to patch before ZDI publishes them. The contest also saw Samsung's Galaxy S26 compromised in all seven attempts against it, alongside successful hacks of OpenAI's Codex, Oracle's Autonomous AI Database, printers from Lexmark, Canon and Brother, smart home gear including Sonos Era 300 and Philips Hue Bridge Pro, and Garmin's Index BPM, with total payouts exceeding $1.2 million across the event.

Citrix has issued fixes for a critical memory overflow bug, tracked as CVE-2026-107406, affecting NetScaler ADC and NetScaler Gateway. The flaw, rated 9.5 on the CVSS scale, can lead to remote code execution or denial-of-service when a device is set up as a SAML identity provider or service provider, including Secure Private Access Hybrid deployments built on NetScaler. Citrix says there's currently no sign of in-the-wild abuse and credits researchers from JPMorgan Chase's XOR Team along with Maxim Suhanov for the find. Affected builds span multiple 14.1 and 13.1 branches (plus FIPS variants); customers should move to NetScaler ADC/Gateway 14.1-73.46, 13.1-64.29, 14.1-73.46 FIPS, or 13.1-37.283 FIPS/NDcPP or later. This comes alongside separate NetScaler vulnerabilities — CVE-2026-88771, CVE-2026-88772, and CVE-2026-88779 — that attackers are already actively exploiting.

Citrix has issued an urgent advisory about a critical flaw affecting NetScaler ADC, NetScaler Gateway, and Secure Private Access Hybrid deployments built on NetScaler. The vulnerability, CVE-2026-107406, scores 9.5 on CVSS and stems from a memory overflow that could allow an attacker to trigger remote code execution or crash the appliance through denial-of-service, specifically on devices set up as SAML SP or SAML IdP under certain configurations. Citrix says it has no evidence of active exploitation so far, but is pressing customers to upgrade immediately to the fixed releases—NetScaler ADC and Gateway 14.1-73.46, 13.1-64.29, 14.1-73.46 FIPS, or 13.1.37.283 for the FIPS/NDcPP branches. This disclosure follows closely on the heels of another NetScaler zero-day, CVE-2026-88779, and two previously exploited zero-days, CVE-2026-88771 and CVE-2026-88772, which were used in attacks against government, financial, education, legal, and professional services targets.

Following last month's record-breaking batch of 973 CVE fixes from Microsoft, analysts expect October's Patch Tuesday to stay large, though likely not surpass that peak. A recent optional update, KB5002907, meant to refresh outdated Microsoft 365 installations instead caused Office 2016 and Office 2019 to be deactivated or removed on many machines, forcing Microsoft to pause the rollout — a reminder that these out-of-support suites still receive sporadic patches that aren't fully vetted against current builds. Apple has already shipped macOS 27 Golden Gate and closed out a zero-day tracked as CVE-2026-86950 across its operating systems on September 29, while Windows 11 version 26H2 has also rolled out, sharing a kernel with 24H2 and 25H2 for easier upgrades. Looking ahead, expect fresh Chrome updates, likely Firefox or Thunderbird patches, and a new Adobe release cycle covering Premiere, Content Credentials, Substance 3D Modeler, Bridge, and Connect, alongside Microsoft's planned fixes for recent Excel clipboard and Windows Backup bugs.

The FBI has taken down seven domains tied to Chinese state-sponsored group Flax Typhoon, cutting off access to two hacking platforms—MicroScan and FishHub—allegedly run by Integrity Technology Group (Integrity Tech), a company with ties to Beijing. MicroScan, a Python-based scanner with over 1,300 pentesting scripts, was used alongside a Mirai-infected device botnet to probe for weaknesses in critical infrastructure, including a South Carolina power company, airports in Japan and Poland, Taiwanese energy firms, and universities, exploiting flaws such as CVE-2015-3306, CVE-2016-3081, CVE-2014-6278 and CVE-2021-22205. FishHub, meanwhile, enabled spear-phishing and follow-on malware deployment, giving attackers remote access and the ability to exfiltrate data—investigators found stolen files from more than 20 organizations on an associated server. The FBI, CISA, NSA and allied agencies jointly released an advisory with indicators of compromise and urged organizations to patch affected systems, limit internet-exposed services, and enable multifactor authentication, marking the latest US action against Integrity Tech following a 2024 botnet takedown and sanctions from the UK and EU.

The FBI, alongside agencies from six other countries, disclosed a joint advisory on October 8 detailing a long-running Chinese espionage operation tied to Integrity Technology Group, a Beijing-linked firm already sanctioned by the US and UK. Since at least January 2021, hackers associated with the company have broken into government, law enforcement, healthcare, religious and other organizations across Southeast Asia, Africa, and North America—plus US critical manufacturing, education and IT sectors—exploiting eight specific flaws (including CVE-2014-6278, CVE-2019-11510, CVE-2021-22205, and CVE-2023-22894) through tools like the custom MicroScan scanner and open-source utilities Nmap, masscan and WPScan. After breaching networks via password spraying with the EBurst tool, fake login pages, or vulnerable web services, the attackers used SoftEther VPN for persistence, DCSync techniques to harvest credentials, and custom bots (Curlc4, office-cli) to siphon Exchange and Microsoft 365 mailboxes via EWS. Most notably, investigators found the group operates a web application granting unnamed third parties direct access to the stolen email content, raising concerns about a broader data-brokering ecosystem; the advisory includes 39 pages of indicators of compromise and recommends patching affected software, enforcing MFA, and monitoring for unusual Active Directory replication.

Cisco Talos' latest Threat Source newsletter opens with a personal essay from new contributor Pierre Cadieux about balancing patching with operational risk, tying into Cybersecurity Awareness Month. The main research item, from Talos' CAIRN project, reveals that malware authors are increasingly inserting natural-language text into their code specifically to mislead AI-assisted malware analysis — a technique Talos now categorizes as "A3: AI-Analysis Evasion." Tracked over 18 months, these tricks range from basic comments telling an AI to disregard a file to sophisticated "template spraying" aimed at specific large language models; Talos found such prompt injections sway an AI's verdict in the attacker's favor roughly 35% of the time, and some A3 families, like MANTLEMAZE, combine this with vulnerable-driver abuse to kill EDR at the kernel level. The issue is significant because any text pulled from a malware sample must be treated strictly as evidence rather than as a trusted instruction within automated detection pipelines, and defenders can use the presence of imperative language aimed at analysis tools as a reliable red flag. The roundup also covers other news, including new Citrix NetScaler zero-day reports (CVE-2026-88779), a breach exposing 8 million Danish citizens' records, Google pausing parts of its open-source bug bounty program, Warlock ransomware hitting Spanish and Portuguese organizations, and the US Senate passing a healthcare cybersecurity bill following the Change Healthcare breach.