Security news, decoded.

JPCERT/CC has warned that a wave of personal data breaches hitting Japanese organizations stems from abuse of mobile app management APIs and exploitation of a critical Metabase SQL injection flaw, CVE-2026-72898 (CVSS 10.0). Attackers reverse-engineer smartphone apps to find hidden API endpoints, exploit internal APIs to escalate privileges or create rogue accounts, and reuse stolen API keys from other breaches; some also hit vulnerable admin panels and known software bugs. Security firm Macnica counted 119 similar incidents in Japan so far in 2026, up sharply from 84 in 2025, including major leaks at Times Car (around 8 million accounts combined) and the Yakiniku King app (over 10.7 million records). Metabase has urged users to upgrade to its listed minimum safe releases (e.g., 0.63.13, 0.62.16, 0.61.18) since attacks on the unpatched flaw continued even after the August fix, and Japan's privacy regulator has separately issued guidance on API-related data leaks.

Cisco has released security updates addressing 35 vulnerabilities across several of its product lines, with more than a dozen rated critical. A Meraki hardening update bundles multiple memory-corruption issues, such as buffer overflows and out-of-bounds writes, under seven CVEs, the most severe being CVE-2026-76464. In License On-Prem, Cisco closed eight flaws, five of them critical, including CVE-2026-20328 (unauthorized access) and CVE-2026-76454 (denial of service), both exploitable without authentication, alongside CVE-2026-76482, CVE-2026-76480 and CVE-2026-76483 covering missing authentication, bad cryptographic signature checks, and poorly protected credentials. NX-OS received the largest batch of fixes, 14 vulnerabilities including seven critical ones: CVE-2026-76455 and CVE-2026-76459 group several improper access control and out-of-bounds write bugs, while CVE-2026-76471 and CVE-2026-76465 let unauthenticated remote attackers run code as root or trigger a DoS; three more issues (CVE-2026-76485, CVE-2026-76486, CVE-2026-76501) only affect Nexus 3000 and 9000 switches with NGOAM enabled. Cisco also patched three critical flaws in Application Policy Infrastructure Controller (CVE-2026-76498, CVE-2026-76499, CVE-2026-76500) tied to access control, OS command injection and memory corruption, plus a publicly disclosed high-severity SSRF bug in Finesse (CVE-2026-20362). The company says none of these issues are currently being exploited in the wild.

Cisco has published fixes for five critical vulnerabilities in NX-OS, the operating system running its Nexus data center switches, which could let an attacker execute code with root privileges or at least crash the device and trigger a reload. The bugs affect Nexus 3000 and 9000 Series switches running in standalone NX-OS mode, and only trigger when the NX-API, Next Generation OAM (NGOAM), or MPLS OAM features are enabled on the device. CVE-2026-76471 stems from weak input validation in NX-API and is reachable via a crafted HTTP request, while CVE-2026-76485, CVE-2026-76486 and CVE-2026-76501 involve flawed IP traffic checks tied to NGOAM (with CVE-2026-76486 additionally requiring SRv6 or NV Overlay, and CVE-2026-76501 requiring SRv6). CVE-2026-76465 concerns improper handling of MPLS echo-request packets when MPLS OAM is turned on, though switches with Silicon One ASICs are immune. Nexus 7000 devices and Nexus 9000 units in ACI mode are not impacted. Cisco says the flaws were found internally, with no evidence of public disclosure or active exploitation, and urges customers to update to a fixed NX-OS build, disable the unused features, or apply the temporary Live Protect shield while patching is scheduled.

Hours after security firm WatchTowr released a technical breakdown of CVE-2026-21589, threat actors began probing Atlassian's self-hosted Data Center products for the flaw. Atlassian had disclosed the bug on October 5 with a CVSS score of 9.3, affecting Bitbucket, Confluence, Jira Software, Jira Service Management, Bamboo, Crowd, Crucible and Fisheye, and had already shipped fixes for every impacted version. The vulnerability allows an unauthenticated remote attacker to pull specific files from the web root as long as the exact file name and path are known, though it cannot be used to browse directories. WatchTowr found the root cause in a shared library and showed that when Jira is linked to Crowd, an attacker could read a config file holding Crowd credentials in plain text — enough, in their words, to effectively take over by creating a rogue admin account. Monitoring firm Previdian recorded roughly 190 exploitation attempts from 32 IPs across 10 countries within two days of the disclosure, though CISA has not yet listed the bug in its Known Exploited Vulnerabilities catalog. Affected organizations are urged to apply the available patches immediately or, failing that, isolate instances from the internet and use Atlassian's suggested firewall and rewrite rules.

SonicWall and Splunk both released security updates this Wednesday addressing a batch of critical and high-severity flaws, some of which allow remote code execution. SonicWall patched four bugs in its SMA1000 appliances, advising immediate upgrades to versions 12.5.0-03082 and 12.4.3-03670. The worst of these, CVE-2026-102255, carries a maximum CVSS score of 10 and stems from an unauthenticated SSRF issue tied to an unintended alternate access route, letting an attacker make the appliance send requests on their behalf to reach internal services. The remaining fixes cover two high-severity RCE-capable issues and one medium-severity XSS flaw; SonicWall says there's no known in-the-wild exploitation so far, and clarified that SSL-VPN on its firewall line is not impacted. Separately, Splunk pushed out fixes for numerous vulnerabilities across Splunk Enterprise, its MCP Server, and the Add-on for Amazon Web Services. Three critical-severity bugs in Splunk Enterprise could enable arbitrary command execution, unauthorized access, or code injection, while a medium-severity MCP Server flaw could let an authenticated user redirect API requests to a server they control. Splunk also addressed several third-party package vulnerabilities affecting Enterprise and the AWS add-on, with full details published on its advisories page.

Security firm Patchstack has detected active in-the-wild attacks exploiting a critical unauthenticated file upload vulnerability in the Bricksforge plugin for WordPress, a popular extension for Bricks Builder. Tracked as CVE-2026-85097 with a maximum CVSS score of 10.0, the flaw lets attackers bypass MIME-type checks using image/PHP polyglot files and manipulate the temporaryFileUploads parameter's url field to drop and execute arbitrary PHP code on the server, achieving full remote code execution. The bug affects all Bricksforge versions up to and including 3.1.8.9, and is fixed in version 3.1.8.10. Patchstack first observed exploitation attempts on October 7, 2026 at 21:47 UTC, mostly hitting the plugin's REST API form-submission endpoint, with attackers probing numerous extension and encoding tricks from dozens of distributed IP addresses; site owners are urged to update immediately or apply a virtual-patching rule while they do.

Black Lotus Labs has uncovered a cryptomining botnet called PoeLLM that retrieves its command-and-control address from a poem hidden in a GitHub repository, with each of the poem's words decoded into numbers that form the C2's IPv4 address. Tracked under the name Canto Incognito and attributed to a likely Italian-speaking, financially motivated actor active since April 2026, the campaign has compromised more than 3,400 servers, mostly in the US and Western Europe, by exploiting exposed instances of LiteLLM, Ollama, Gotenberg and Gitea, as well as Ivanti Sentry (CVE-2026-10520). Researchers first spotted the malware while investigating that Ivanti flaw, and later linked infections to a command-injection bug in LiteLLM tracked as CVE-2026-42271. Once a device is breached, PoeLLM deploys XMRig and Iron miners tied to the Kryptex pool, while also offering remote-shell access and scanning/exploitation tools that let infected machines hunt for new victims; the operator has reportedly begun testing SSH brute-forcing as well, and Black Lotus Labs says it is now blocking traffic to the botnet's infrastructure.

Florida, Iowa, Montana, and Nebraska have filed consumer-protection lawsuits against TP-Link Systems, following a similar Texas case from February, alleging the company exaggerated the security of its routers and downplayed its links to China. The complaints cite claims from TP-Link's HomeShield marketing and reference congressional testimony tying TP-Link devices to the Volt Typhoon and Flax Typhoon campaigns, as well as separate botnet and Russian hacking incidents involving the company's hardware. As supporting evidence of ongoing security weaknesses, the lawsuits reference five vulnerabilities — CVE-2025-30237 through CVE-2025-30241 — affecting TP-Link's Aginet-branded mesh systems, routers, and modems, which TP-Link disclosed in August and which SEC Consult researchers detailed this week; the most critical, CVE-2025-30237, lets an unauthenticated attacker on the local network gain full administrative control via an authentication bypass, while others involve privilege escalation, root-level command injection, hardcoded encryption keys exposing credentials, and a USB-based physical attack vector. TP-Link says patches are being rolled out through ISPs, with the process extending into 2026, and has called the lawsuits baseless, even as a coalition of 21 state attorneys general separately pushes the FCC to deny the company's pending approval to sell new router models in the US.

Cisco Talos describes a growing malware trend dubbed "AI-analysis evasion," in which samples embed plain-language text aimed at manipulating automated AI tools used during malware triage rather than at hiding from traditional binary analysis. Tracked through Talos' CAIRN project across four families — FRUITSHELL, PLOTSAFE, HOLLOWCLAD and MANTLEMAZE, covering 84 samples collected between January 2025 and July 2026 — the technique ranges from a simple comment telling an LLM the file is harmless, to templated and randomized versions, to payloads formatted to mimic chat-prompt syntax from multiple AI vendors, and even fabricated warnings meant to intimidate a model into refusing to help an analyst. Talos' testing found the crudest, cheapest method (direct instructions telling the model to ignore the file) swayed verdicts in roughly 35% of trials, while more elaborate approaches had inconsistent or even counterproductive effects, often increasing suspicion instead. No CVE or active exploitation is tied to this report beyond a noted BYOVD abuse of CVE-2015-2291 in the MANTLEMAZE loader chain; Talos stresses defenders should treat any embedded text in a sample as evidence, never as an instruction, when building AI-assisted detection pipelines.

Cisco Talos's vulnerability research team has published details on seven flaws discovered across products from Adobe, Apple, Foxit, and Microsoft, all already fixed by the vendors under Cisco's coordinated disclosure process. In Photoshop's installer (version 2.11.0.30), CVE-2026-48388 lets an attacker swap in a malicious file to gain elevated privileges, while macOS CoreWLAN (26.3.1) carries an information-leak bug reachable via a specific API call sequence, tracked as TALOS-2026-2376. Foxit Reader 2026.1.1.36485 is hit by two separate issues — a JavaScript checkbox handling flaw (CVE-2026-57256) and an Array object use-after-free (CVE-2026-91799) — both exploitable through booby-trapped PDF files to achieve remote code execution. On Windows, Talos disclosed an information-disclosure bug in the NETIO.sys driver (CVE-2026-50475) plus three issues in the Cloud Files Mini Filter Driver and tcpip.sys (CVE-2026-58613, CVE-2026-80093, CVE-2026-49177) that can be triggered through crafted API calls or network packets to cause privilege escalation, type confusion, or out-of-bounds reads. Snort rule updates covering exploitation of these bugs are available from Snort.org.

Security teams at CloudSEK and Checkmarx have uncovered a supply-chain operation, dubbed MALFEX, built around a dozen npm packages published since August 2023 by what appears to be a single operator. Eight of those packages — including tlxbnhd, tldriver, mxdriver, img-to-native, native-runner, function-flag, function-color and cdn-img-fetch — were found to be malicious and have racked up 40,767 total downloads, with function-flag alone accounting for 37,419 of them. Depending on the package, victims on Windows end up with a loader for the Go-based Overlord RAT (which hides its C2 address inside Solana blockchain transactions), a Node.js stealer called movinlike that grabs data from Discord, browsers, Telegram and crypto wallets, or a downloader chain that fetches a Go binary before dropping the stealer; three packages (function-flag, function-color, cdn-img-fetch) remain published on npm. Researchers note Overlord RAT has also surfaced in a WordPress exploitation wave tied to CVE-2026-63030 and CVE-2026-60137, and in a fake-Zoom-installer campaign on macOS linked to tactics resembling the North Korea-aligned cluster UNKDeadDrop, though the npm operator is believed to be a separate, Portuguese-speaking individual likely based in Brazil.

SonicWall has shipped hotfixes for four vulnerabilities in its SMA1000 remote-access appliances, the most severe being a maximum-severity, pre-authentication SSRF bug in the WorkPlace login portal, tracked as CVE-2026-102255 and scored 10.0 on CVSS. The flaw stems from an unintended access route that lets an unauthenticated attacker reach internal appliance functions and carry out unauthorized actions, though SonicWall says it has found no sign of active exploitation so far. The remaining three bugs — CVE-2026-102256 (command injection enabling RCE), CVE-2026-102257 (a Zip Slip path-traversal issue in the Appliance Management Console), and CVE-2026-102258 (stored XSS in AMC) — all require prior login. All four affect SMA1000 models 6210, 7210, and 8200v running platform-hotfix versions 12.4.3-03526 or earlier, and 12.5.0-02952 or earlier; fixes land in 12.4.3-03670 and 12.5.0-03082 respectively, available via the MySonicWall portal. This marks the third unauthenticated 10.0-rated SSRF flaw SonicWall has patched in WorkPlace this year, after two earlier pairs in July and September that the company confirmed were actively exploited in real attacks.

Security researchers at JFrog have disclosed a critical, still-unpatched vulnerability in LMCache, the open-source caching layer used to accelerate LLM inference servers like vLLM. Tracked as CVE-2026-105192 and rated 9.8 out of 10, the bug lets an unauthenticated attacker execute arbitrary code on the cache server simply by sending a crafted network message. The issue lies in LMCache's multiprocess mode, where worker processes communicate with a standalone cache server over ZeroMQ. That server deserializes certain incoming messages using Python's pickle format before validating them, so a malicious payload can trigger code execution with the privileges of the LMCache process — which runs as root in the project's official container images. Exploitation requires the server to be bound to a network-reachable address rather than localhost, a configuration used in LMCache's own example Kubernetes deployment. The flaw affects LMCache versions 0.3.9 through 0.5.5, as well as the 0.5.6 release candidates and the development branch, with no fixed version currently available. JFrog recommends avoiding routable addresses for the multiprocess server and restricting network access until a patch is released. Separately, six unconfirmed security reports against LMCache surfaced on GitHub, and a related, already-patched vLLM denial-of-service bug (CVE-2026-105756) was fixed in version 0.30.0.

A new cryptomining botnet called PoeLLM has been hijacking poorly secured AI infrastructure to mine cryptocurrency and spread further through networks. According to Lumen's Black Lotus Labs, the malware has already infected more than 3,400 machines, with up to 800 active in a single day, mostly running exposed tools like LiteLLM, Ollama, Gotenberg, and Gitea, alongside some Ivanti Sentry instances in the US and Western Europe. Unusually, the malware hides its command-and-control address inside a poem published on GitHub, converting specific words into an IP address through a built-in dictionary — the operators have already swapped the poem 11 times to rotate infrastructure. Besides deploying XMRig and Iron miners, the malware scans for and exploits CVE-2026-42271 in LiteLLM's MCP test endpoints, which researchers at Horizon.ai showed can be chained with CVE-2026-48710 for unauthenticated remote code execution, making prompt patching and limiting internet exposure of AI deployments essential.

Atlassian issued patches on October 5th for an arbitrary file access flaw tracked as CVE-2026-21589, affecting several of its products including Jira, Confluence and Bitbucket. The bug stems from how these applications sanitize path traversal attempts by substituting slashes with a "::" sequence, which under certain conditions can be reversed by attackers to escape the intended resource directory and read files such as web.xml inside WEB-INF, exposing configuration data. A detailed technical breakdown and working proof-of-concept requests were published by Watchtowr shortly after the patch release. SANS ISC reports that since October 6th its honeypots have recorded exploitation attempts using URL patterns matching those PoCs, targeting Jira, Bitbucket and Confluence webresource endpoints. The requests all appear to originate from a single actor operating through a cluster of Digital Ocean-hosted IP addresses, suggesting automated mass scanning is already underway against unpatched instances.

Just a day after Atlassian shipped fixes for a critical arbitrary file access bug (CVE-2026-21589) in its self-managed Data Center products, and hours after watchTowr researchers published technical analysis, threat intelligence firm Previdian reported seeing real exploitation attempts hitting its honeypots, along with a list of attacker IP addresses. The flaw impacts every version of Bitbucket Data Center, Confluence Data Center, Jira Service Management Data Center, Jira Software Data Center, Bamboo Data Center, Crowd Data Center, Crucible and Fisheye, and stems from a routing defect in the shared library atlassian-plugins-webresource.jar, where a double-colon sequence bypasses slash-stripping protections and enables path traversal to read arbitrary files. watchTowr demonstrated that this can expose Crowd's plaintext credentials file, letting attackers authenticate to Crowd's SSO system, create accounts, and escalate to Jira administrator privileges. Atlassian is urging immediate patching, or isolating vulnerable instances from the internet and reviewing access logs for signs of compromise in the meantime.

A Jordanian teenager known online as "Rey," allegedly the current leader of the data-theft and extortion gang ShinyHunters, has been detained and is reportedly cooperating with the FBI to expose other members. At the time of his arrest he was in the middle of extorting Jeppesen ForeFlight, an aviation-software unit that Boeing sold off to Thoma Bravo last year, and both companies have confirmed the extortion attempt while saying operations were not impacted. Rey previously claimed credit for breaching the FBI and the Cl0p ransomware group by exploiting CVE-2026-35273, a flaw in Oracle's PeopleSoft platform that ShinyHunters began abusing as a zero-day in June and that researchers at Mandiant and Google's Threat Intelligence Group say was used to mass-harvest data from dozens of organizations. The case highlights how ShinyHunters has effectively become a franchised brand run by shifting, loosely affiliated operators rather than its original founders.

German authorities have taken custody of a 28-year-old Russian national suspected of being a core member of the Qilin ransomware operation, after his arrest in Osaka, Japan in May and his handover on October 2. He is wanted in Germany over a September 2024 breach of a logistics company, where he allegedly encrypted systems and extorted the victim for more than $160,000 in cryptocurrency. Qilin, also known as Agenda, has run as a ransomware-as-a-service operation since August 2022 and is tied to major incidents including the 2024 Synnovis attack that disrupted London NHS hospitals, the 2025 breach of Asahi Group affecting roughly 2 million people, and roughly 400 victims posted to its leak site this year, including Lee Enterprises, Inotiv and the US ATF. The arrest marks a rare law-enforcement success against one of the most active and damaging RaaS groups currently operating.

Threat actors have begun exploiting CVE-2026-21589, a critical unauthenticated file-access bug affecting self-hosted Atlassian products such as Jira, Confluence, and Bitbucket, among eight impacted Data Center/server tools. Security firm Previdian says its honeypots picked up active scanning and exploitation attempts within just two hours of researchers at watchTowr publishing a detailed writeup and public proof-of-concept exploit. The flaw stems from a shared web-resource component that mishandles double-colon sequences, letting attackers craft directory-traversal requests through plugin endpoints to read restricted files without logging in. In Crowd-integrated setups, this can expose plaintext credentials that let attackers escalate to full Jira administrator access via Crowd's API. A Nuclei template is now circulating too, making mass scanning trivial, so researchers expect attack volume to keep climbing. Atlassian customers running affected self-hosted instances should install the vendor's patches immediately or apply the listed network and web-server mitigations without delay.

Attackers started probing a critical unauthenticated file-access flaw in Atlassian's Data Center software just two hours after researchers at watchTowr published technical exploitation details. The bug, CVE-2026-21589 (CVSS 9.3), lets an unauthenticated actor fetch specific files from the web application root of Bitbucket Data Center, Confluence Data Center, Jira Software and Jira Service Management Data Center, Bamboo Data Center, Crowd Data Center, Crucible and Fisheye, provided they already know the exact file path. Security firm Previdian logged 15 exploitation attempts from three IPs in Japan and the US hitting its honeypots, and warned that in Crowd and Jira deployments the technique can expose crowd.properties credentials, enabling attackers to create rogue accounts with full Jira Administrator rights. Atlassian has shipped patched builds for each affected product and urges customers to upgrade immediately or apply interim mitigations such as WAF rules, Tomcat RewriteValve blocks, or urlrewrite.xml changes while a public Nuclei template is expected to accelerate mass scanning.

SonicWall has issued hotfixes for a maximum-severity server-side request forgery bug affecting its SMA1000 series secure remote access gateways, specifically the 6210, 7210, and 8200v models. Tracked as CVE-2026-102255, the issue lies in the Appliance WorkPlace interface and arises from an unexpected alternate access path, allowing an unauthenticated remote attacker to force the device into making requests on their behalf and reach internal functions without needing much skill or access. The SMA 100 line and SonicWall firewall SSL-VPN services are not impacted. SonicWall says it has no evidence of active exploitation so far, but given that SMA1000 appliances have repeatedly been targeted in zero-day attacks this year—including ransomware-linked campaigns flagged by CISA—customers are strongly urged to apply the fix promptly, especially since Shadowserver still counts over 400 internet-facing SMA1000 devices.

Google has shipped Chrome 155, a security update closing 247 vulnerabilities, among them four critical use-after-free bugs affecting the Chromecast, Browser, Navigation, and Track components (CVE-2026-106382, CVE-2026-106197, CVE-2026-106358, and CVE-2026-106347). One of these was found internally by Google, while researcher Xinyang Ge reported the other three, partly with the help of AI-assisted bug hunting. The release also addresses 53 high-severity issues and 190 lower-risk medium/low findings, with Google paying out around $33,000 in bounty rewards for externally reported bugs, though many payouts remain undisclosed. No evidence of active exploitation has been reported for any of the fixed flaws, but the volume and severity mix make prompt updating to version 155.0.8059.39/.40 (Windows/macOS) or 155.0.8059.39 (Linux) advisable.

SonicWall has released fixes for four vulnerabilities in its SMA 1000 series SSL VPN gateways, the most serious of which, CVE-2026-102255, lets an unauthenticated remote attacker abuse the appliance's Work Place interface to trigger an SSRF and trick the device into making requests to internal, trusted endpoints — effectively performing actions reserved for authenticated users or admins. The vendor says there is no current evidence of in-the-wild exploitation, but notes the risk given that SonicWall SMA appliances have repeatedly been targeted as zero-days in the past, including two other pre-auth SSRF bugs earlier in 2026 (CVE-2026-15409 and CVE-2026-83548). The other three flaws — CVE-2026-102256 (post-auth OS command injection), CVE-2026-102257 (path traversal) and CVE-2026-102258 (XSS) — require authenticated access, with the latter two needing admin credentials. The issues affect SMA 1000 models 6210, 7210 and 8200v (physical and virtual); SonicWall firewalls and the discontinued SMA 100 line are not impacted. Administrators should update to hotfix firmware versions 12.4.3-03670 or 12.5.0-03082 and higher.

Anthropic has broadened its Cyber Verification Program, letting vetted security professionals use Claude models—including Opus 5.5, Sonnet 5.5 and Mythos 5.1—with fewer built-in restrictions for defensive and authorized offensive security work. The move follows results from Project Glasswing, which Anthropic says identified roughly 129,000 verified software vulnerabilities between April and July 2026, plus another 5,500 found via open-source scanning through October, with over 33,000 rated critical or high severity. The program now offers three tiers—Defense Access, Red Team Access, and a highly restricted Specialized Access—each unlocking different levels of safeguard relaxation for incident response, malware analysis, and penetration testing. Separately, outside research noted that only a small fraction of flaws surfaced by Anthropic-linked efforts have actually been exploited in the wild, underscoring that AI-assisted discovery doesn't always equate to real-world risk.

Atlassian has issued fixes for a critical flaw, CVE-2026-21589 (CVSS 9.3), that lets unauthenticated attackers read specific files within the web root of eight products, provided they already know the exact file name and path. All versions of Bitbucket Data Center, Bamboo Data Center, Crowd Data Center, Crucible, Confluence Data Center, Fisheye, Jira Service Management Data Center, and Jira Software Data Center are affected, with fixed releases now available for each (for example Bitbucket 10.5.1, Confluence 10.2.19, and Jira 11.3.12). Atlassian and security firm WatchTowr say there's no sign of active exploitation yet, but warn that similar bugs have previously been weaponized by ransomware crews and APTs, and that Crowd deployments using SSO are especially risky since plaintext authentication data sits in a predictable location that could let attackers forge admin accounts. Administrators are urged to patch self-hosted instances immediately or, if that's not possible, cut off public internet access and apply the vendor's WAF mitigations in the meantime.

X.Org has released fixes for a dozen security issues affecting the X server and Xwayland, bundled into xorg-server 21.1.25 and xwayland-24.1.14. Nine of the flaws could allow arbitrary code execution, while the remaining three may crash the server or leak information. Most of the issues (ten out of twelve) require an already-authenticated X client to trigger them, though CVE-2026-93524 and CVE-2026-93536 don't list that requirement. Eleven of the twelve bugs hit both the X server and Xwayland, with CVE-2026-93522 — a heap buffer overflow in Glamor's CopyArea routine on GPU-accelerated setups — affecting Xwayland only. The root causes vary: seven are buffer overflows or out-of-bounds writes, three are use-after-free conditions, one is a double-free, and one is an out-of-bounds read. Two bugs rely on commonly enabled extensions — CVE-2026-93515 needs Present and SYNC active, while CVE-2026-93519 requires XFIXES, XTEST, and more than 100 live pointer barriers. Two fixes also close gaps left by earlier patches: CVE-2026-93520 completes a fix started in commit a3171732d, and CVE-2026-93521 mirrors a previously patched issue in RRChangeOutputProperty that had only been addressed for the RandR output path, not the provider path. Administrators running X server or Xwayland should check their installed version against 21.1.25 and 24.1.14 respectively and apply the update, since there is no indication the flaws are being exploited yet but code execution risk makes patching a priority.

Security researchers at Patchstack have detected active exploitation of stored XSS flaws in two popular WordPress plugins—Ninja Forms and WPC Product Bundles for WooCommerce—to plant backdoors and hidden administrator accounts. The bugs, CVE-2026-94504 (Ninja Forms, versions 3.15.3 and earlier) and CVE-2026-93836 (WPC Product Bundles, versions 8.6.6 and earlier), both require an authenticated session but are being abused through the same malicious script served from imgcdn1[.]com, suggesting a single threat actor behind both campaigns. The attack injects JavaScript into form submissions or order data that, once viewed by a logged-in admin, installs a fake plugin disguised as "WP Smart Thumbnails" and creates a privileged admin account hidden from the normal WordPress user list, along with a secret login URL and an unauthenticated file manager for persistence. Even after the rogue plugin is deleted, these hidden access points can survive via auxiliary backdoor plugins with falsified timestamps, so site owners are urged to update to Ninja Forms 3.15.4+ or WPC Product Bundles 8.6.7+ and actively check for signs of prior compromise, since patching alone won't remove an existing infection.

WordPress released version 7.1.3 on 6 October 2026, closing seven security holes alongside four bug fixes. Unlike the previous 7.1.2 update, which patched a single critical unauthenticated local-file-inclusion-to-RCE bug, this release addresses a mix of lower-severity issues: an unauthenticated stored XSS in pending comments that requires a moderator click, a feed leak exposing comments on private/unpublished posts to anyone, a denial-of-service in WPHttp::makeabsoluteurl() reachable by Contributor accounts, a second-order SQL injection in WXR exports triggerable by admins exporting content with a malicious thumbnailid, a privilege bug letting Authors sticky posts, an XSS via Imgur oEmbed embeds, and a hook-naming collision that could let a crafted post status/type fire unrelated action hooks like deletepost. None of these require urgent emergency action, but administrators should update promptly and review which users hold Contributor-or-higher roles, since several flaws depend on that access level. Sites using automatic updates will receive 7.1.3 automatically, and fixes have been backported through at least the 6.6 branch, though versions 4.7–6.5 remain unpatched for now.

Rapid7 researchers have uncovered Linux malware families hitting telecom and network edge appliances in South Korea and Taiwan that disguise themselves as trusted email-security software to avoid detection. In South Korea, a new BPFDoor variant and a related BPF Rekoobe build masquerade as processes belonging to the SpamSniper anti-spam product, while one sample also mimics Oracle's PMON process naming. In Taiwan, attackers deployed a previously undocumented implant called AVERAT, dropped through an ELF loader hidden in a ShareTech appliance directory and using encrypted SMTP traffic over port 25 for command-and-control. Both toolsets show how threat actors are refining BPF-based backdoors and abusing the trusted status of email-security appliances to blend malicious traffic into legitimate network noise, echoing earlier abuse of Barracuda ESG flaws tracked as CVE-2023-2868 and CVE-2023-7102.

Atlassian has disclosed CVE-2026-21589, a critical vulnerability affecting several self-hosted Data Center products, including Jira, Confluence, Bitbucket, Jira Service Management, Bamboo, Crowd, Crucible, and Fisheye. The flaw lets an unauthenticated attacker retrieve specific files from an application's web root, though exploitation requires already knowing the exact file name and path, meaning directory listing is not possible. Fixed versions are now available for each product, and cloud deployments have already been patched automatically by Atlassian, while self-hosted administrators are urged to update every cluster node immediately. Atlassian says it has no current evidence of active exploitation but recommends restricting network exposure, applying WAF or proxy rules, and reviewing logs for traversal attempts as temporary safeguards until patching is complete.

Atlassian issued an out-of-band security advisory for CVE-2026-21589, a critical pre-authentication arbitrary file read flaw affecting almost its entire Data Center product line: Jira Software, Jira Service Management, Confluence, Bitbucket, Bamboo, Crowd, Crucible and Fisheye, across all versions prior to the newly released fixes. Researchers at watchTowr traced the root cause to a shared library, atlassian-plugins-webresource, where a resource-routing mechanism meant to escape slashes as double colons (::) can be abused to build path-traversal sequences and reach files inside the WEB-INF directory of any affected application. While this alone exposes configuration and server files, the researchers showed that in deployments integrated with Atlassian Crowd, reading the crowd.properties file leaks plaintext application credentials that let an attacker log into Crowd's management API directly, create a new user, and grant it administrator privileges — turning a file-read bug into full administrative takeover. Given the hundreds of thousands of internet-facing Confluence and Jira instances alone, organizations running any of the listed on-premises Atlassian products should apply the vendor's fixed versions immediately.

A new guide outlines eight security controls that managed service providers should verify before trusting any RMM platform, built from Acronis's experience securing endpoint management at scale across its Cyber Platform. The checklist stresses testing real outcomes—device discovery, risk-based patching, privileged access separation, alert noise reduction, safe automation, incident response integration, recovery validation, and tenant isolation—rather than relying on vendor feature lists. The piece points to recent incidents as proof that RMM tooling itself is part of the attack surface: N-able's N-central platform needed an emergency hotfix for the maximum-severity pre-auth RCE flaw CVE-2026-86218 (its fourth fix in five weeks, with about 1,500 servers exposed), while Microsoft SharePoint's "ToolShell" zero-days, CVE-2025-53770 and CVE-2025-53771, were exploited before patches existed and compromised at least 85 on-premises servers. CISA has similarly flagged ransomware groups abusing legitimate RMM tools to pivot into customer networks, underscoring why MSPs need hands-on validation of these controls rather than assuming they are covered by default.

Atlassian has disclosed a critical flaw, tracked as CVE-2026-21589 with a CVSS score of 9.3, that lets an attacker read files from the web root of several Data Center products without any authentication. The issue affects every version of Bitbucket Data Center, Confluence Data Center, Jira Software Data Center, Jira Service Management Data Center, Bamboo Data Center, Crowd Data Center, Crucible and Fisheye, though exploitation requires the attacker to already know the exact name and path of the targeted file, since directory listing isn't possible. Cloud versions of these products have already been fixed and Atlassian says it found no sign of in-the-wild abuse there, but on-premise Data Center customers are urged to upgrade immediately to the patched releases—such as Bitbucket 9.4.26/10.2.8/10.5.1, Confluence 9.2.26/10.2.19, Jira 10.3.26/11.3.12, Bamboo 10.2.24/12.1.12, Crowd 6.3.7/7.0.3/7.1.7/7.2.4, and Crucible/Fisheye 4.9.15—or at minimum isolate internet-facing instances and apply the temporary mitigations Atlassian has published while reviewing logs for signs of compromise.

Researchers have published a proof-of-concept technique that lets a booby-trapped Calc spreadsheet execute arbitrary code the moment it's opened, bypassing the usual macro-security prompt in both LibreOffice and Apache OpenOffice. The trick abuses a legitimate "database range" feature: the spreadsheet points to an external ODB database file, which in turn references a remote JDBC driver JAR that gets downloaded and loaded automatically, effectively running attacker-supplied Java code as soon as Java support is enabled. LibreOffice has already patched its flaw, tracked as CVE-2026-63277, in the October 5 releases 26.2.5 and 26.8.0, while Apache OpenOffice's equivalent issue, CVE-2026-59265, remains unfixed in all versions through 4.1.16, with a patch expected in 4.1.17. There is no known in-the-wild exploitation yet, but OpenOffice users are advised to disable Java or avoid opening untrusted spreadsheets until a fix ships.

Dell has released a fix for a severe security flaw in its Dell System Update (DSU) utility, a tool used by enterprise administrators to deploy driver, BIOS, and firmware updates on PowerEdge servers. Tracked as CVE-2026-86360 and rated 9.6 on the CVSS scale, the path traversal bug affects all DSU releases before version 2.3.0.0 and could let an attacker without any credentials run code as root over the network, potentially taking full control of both the application and the host system. Dell says it has no evidence of active exploitation but strongly urges customers to upgrade immediately to DSU 2.3.0.0 or later. Alongside this flaw, Dell also patched four other high-severity DSU bugs — CVE-2026-63697 and CVE-2026-71168, which enable remote code execution, and CVE-2026-86361 and CVE-2026-86362, which allow privilege escalation — credited to researchers Ori Gabriel, saltedfish, and Nir Yehoshua of Cipher Security Labs.

Patchstack researchers have tracked a single malicious campaign abusing two unrelated stored Cross-Site Scripting flaws in WordPress plugins — CVE-2026-93836 in WPC Product Bundles for WooCommerce (up to version 8.6.6) and CVE-2026-94504 in Ninja Forms (up to version 3.15.3) — to deliver the exact same JavaScript implant hosted on imgcdn1.com. First seen on October 4-5, 2026, the script hijacks a logged-in administrator's session to install a disguised rogue plugin, open an unauthenticated file manager, create a visible administrator account, and plant must-use plugins that hide a second administrator from the Users screen while generating a magic login link that impersonates the site's original owner. This gives attackers four separate, persistent paths back into the compromised site that survive removal of the original vulnerable plugin. Site owners should update WPC Product Bundles to 8.6.7 and Ninja Forms to 3.15.4 or later, and specifically hunt for hidden admin accounts, rogue mu-plugins, and the fzemer database options rather than relying on the dashboard alone.

Atlassian has disclosed CVE-2026-21589, a critical path traversal vulnerability rated 9.3 out of 10, affecting eight self-hosted Data Center products: Bitbucket, Confluence, Jira Software, Jira Service Management, Bamboo, Crowd, Crucible and Fisheye. An unauthenticated attacker who already knows a file's exact name and path can read that file from the web application's root directory, though they cannot browse the directory's contents. Atlassian has released fixed versions for every affected product and notes its cloud offerings are already patched and show no sign of exploitation so far, while urging self-hosted customers who cannot upgrade immediately to take instances offline or apply temporary WAF, Tomcat RewriteValve, or urlrewrite.xml blocking rules. The company also recommends checking access logs for suspicious path-traversal patterns, since a similar flaw in Jira (CVE-2021-26086) was previously exploited in the wild.

The FBI has cut ties with an Accenture-supplied contractor after determining that a missed security patch allowed the ShinyHunters group to breach a bureau system and steal personal data on thousands of employees. According to Reuters, FBI cyber division official Brett Leatherman said the incident stemmed from a third-party-managed platform where a required fix was never applied, later confirmed to be Oracle PeopleSoft. Mandiant has linked the intrusion to ShinyHunters abusing a bypass of CVE-2026-35273, using URL encoding to slip past a web application firewall rule meant to shield the vulnerable PSEMHUB endpoint. Accenture says it remains committed to supporting the FBI, while the bureau continues pursuing leads tied to ShinyHunters, a group that has already seen two alleged members arrested.

Microsoft Threat Intelligence has uncovered a new ClickFix variant that stashes its malicious script inside the browser cache instead of fetching it remotely, letting attackers get around the roughly 260-character limit of the Windows Run dialog (Win + R). Compromised sites secretly pre-load a script disguised as a PNG image into the browser's cache; victims tricked into pasting a short command then trigger execution of that already-cached content rather than downloading anything new. In the observed chain, a VBScript scans Firefox profile folders for files matching expected byte sizes, copies a matching cache entry into a .vbs file, and runs it via wscript.exe, eventually pulling PowerShell stages that inject code into a legitimate process (timeout.exe) to steal browser and device credentials. No specific CVE is tied to this technique, but it builds on prior cache-smuggling tricks documented since late 2025 and underscores why users should never paste 'fix' commands into Run, Terminal, or PowerShell prompts.

A researcher named Syed Anas Mohiuddin has demonstrated a new attack class, dubbed "protocol pivoting," that abuses the trust model built into the Model Context Protocol (MCP), the emerging standard used by AI agents to communicate inside corporate networks. By injecting malicious instructions into one agent, an attacker can get that agent to relay the harmful commands to another internal agent, which blindly executes them because it inherently trusts the first agent—even when the two use different inter-agent protocols such as Google's Agent-to-Agent (A2A). Mohiuddin found the flaw affecting five organizations with little in common besides MCP adoption, including Google, JP Morgan Chase, Weaviate, Rapid7, and French and US government systems. Rapid7's bug, tracked as CVE-2026-97228 and rated a low 2.7 severity, was patched last month, while Google's more serious issue (CVSS 8) stemmed from its mcp-toolbox for databases failing to validate redirect targets and IP addresses, enabling server-side request forgery; Google fixed it by adding IP allow/block lists and startup-time URL validation. Experts warn the underlying weaknesses are classic injection and SSRF bugs, but the rapid, unvetted rollout of agentic architectures across industries has sidelined basic zero-trust principles, leaving many MCP deployments exposed to this cross-protocol trust abuse.