🔴 SonicWall SMA1000 KEV deadline Sep 5, 🔵 Sangoma Switchvox CVSS 10.0 KEV, 🟠 Cisco Nexus 9000 pre-auth RCE, 🟣 WP Migration SQLi millions of installs, 🟡 Spring Ring Teams IT helpdesk impersonation, 🟢 GitSpawn AI agent RCE vector

CISA set a September 5 remediation deadline for the SonicWall SMA1000 zero-day chain after adding CVE-2026-83548 and CVE-2026-83549 to its Known Exploited Vulnerabilities Catalog. The SSRF-to-RCE pair, both CVSS 10.0, targets internet-facing remote-access appliances across models 6210, 7210, and 8200v. Rapid7's MDR team confirmed active exploitation with attackers chaining the two flaws to achieve unauthenticated command execution from the network perimeter. SonicWall's guidance calls for immediate hotfix deployment, appliance re-imaging, password resets, and TOTP token revocation. With the three-day federal deadline and six sources tracking the story since mid-July, this remains the highest-priority remediation target across the entire landscape.

Horizon3.ai disclosed active exploitation of CVE-2026-9586, a critical SQL injection in Sangoma Switchvox that escalates to remote code execution, and CISA added it to the KEV Catalog the same day. The flaw sits in the /pa HTTP endpoint handling phone notification events, where a PhoneIP parameter is concatenated directly into a PostgreSQL query without parameterization. Honeypots captured exploitation from IP 176.65.148.184 attempting reverse shells across multiple systems in rapid succession. Switchvox is an on-premises enterprise VoIP platform handling phone systems, voicemail, and device provisioning, and the researchers noted that most internet-facing instances have either already been targeted or will be soon. Fixes shipped in Switchvox 8.4.0.2 on July 14.

Cisco patched CVE-2026-20212, a CVSS 9.8 remote code execution flaw in Cisco Nexus 9000 Series switches equipped with Silicon One ASICs. The vulnerability sits in TCP ports 43210 and 43211 accessible through the default Layer 3 VRF instance, letting an unauthenticated attacker connect and send crafted input to achieve arbitrary code execution with root privileges. Root access on a Nexus 9000 means full control of the data-center fabric: configuration manipulation, traffic interception, credential collection, and lateral movement. Cisco also warns that exploitation attempts can crash the S1HAL process, triggering a device reload. This is an emergency-patch scenario for any internet-exposed or management-accessible Nexus 9000 with Silicon One.

A second-order SQL injection in the All-in-One WP Migration plugin (CVE-2026-19949) allows unauthenticated attackers to achieve remote code execution on WordPress sites running versions through 7.109. Discovered by researcher Jack Taylor and reported through Wordfence, the flaw stems from incorrect parsing of escaped backslashes and quotation marks during the database restoration process. The attack chain is notably low-barrier: an attacker plants crafted data via WordPress trackbacks, then waits for an administrator to perform a routine export or import operation. When that happens, the injected SQL executes and leaks the plugin's secret import key through a public comment, enabling a malicious .wpress archive import that achieves full site takeover. With millions of active installs and typically slow patch adoption in the WordPress ecosystem, exploitation in the wild is expected within days.

The Spring Ring campaign targeted more than 150 employees across at least 10 organizations by creating external Microsoft Teams accounts impersonating corporate IT help desks. Attackers used names like "Help Desk," "IT Assistance," and "Network Support" with attacker-controlled .onmicrosoft.com tenants to appear legitimate, then initiated Teams chats and followed up with voice calls posing as IT technicians. In the more sophisticated variant, victims were convinced to launch legitimate remote-support software like Quick Assist, after which attackers performed reconnaissance using whoami /groups and net group /dom. PowerShell was used to download an obfuscated RAT from san-sid[.]c, and in a separate variant NTLM relay techniques captured credential hashes after gaining initial remote access. The campaign ran between January and April 2026, with 10-to-15-minute call durations suggesting patient, persuasive social engineering that bypasses traditional email security gateways entirely.

Manifold Security researchers disclosed GitSpawn, a vulnerability class affecting AI coding agents that automatically execute Git commands when opening a project directory. The attack exploits Git's core.fsmonitor configuration option, which lets a repository define a file-system-monitor helper program read from the repo's local .git/config file. An attacker plants an arbitrary command in a malicious repository's Git config, and when an AI agent opens the project and runs routine background commands like git status or git diff, Git launches the attacker's helper program with zero user interaction. Affected tools include Claude Code, OpenAI Codex, Cursor, Grok Build, Goose, Hermes Agent, and Qwen Code. This is a supply-chain class attack vector where a poisoned repository becomes an RCE trigger the moment any AI coding agent touches it.

ThreatFabric identified StreamRAT, a new Android banking trojan distributed through fake free-TV streaming ads on Meta and TikTok, potentially reaching around 570,000 Meta users with Spain as the primary target. The payload was hosted on a GitHub repository linked to an operator previously associated with the Mirax trojan and GodFather malware. StreamRAT uses a multi-stage approach: a dropper masquerading as a streaming client guides victims through enabling installation from unknown sources and Accessibility Services, then requests VPN permission to route most device traffic through a non-functional connection that excludes only the malware itself. The final payload exploits Accessibility APIs, MediaProjection for screen capture, and HVNC for full remote device control, combining social engineering through piracy lures with persistent device access.

A Chinese-speaking threat cluster linked to Earth Berberoka has compromised Brazilian government and educational web servers to conduct large-scale SEO poisoning and online-gambling fraud, active since mid-2025. Check Point Research identified the attackers deploying custom Apache modules on compromised Linux servers that act as secret reverse proxies, silently forwarding selected requests to attacker infrastructure while the real site continues to function normally for administrators and most visitors. The modules react to specific URL paths including /wps, /bmw, and /card, serving attacker-controlled pages from trusted .gov.br domains to boost fraudulent gambling pages in search rankings. The stealth of this approach is notable: the custom modules do not break the site's normal appearance or functionality, meaning standard website monitoring may miss the infection entirely.

Kaspersky researchers highlighted a growing evasion technique in QR-code phishing where attackers build QR codes directly in emails using HTML table cells or Unicode text characters instead of embedding them as image attachments. Because no image object exists in the message, Secure Email Gateways that rely on image inspection, OCR, and QR decoding fail to detect the malicious URL. ESET's H1 2026 Threat Report places QRCode/Phishing among the top email threat detections at 4.1% of the top 10. The scanning and browsing happens on a personal mobile device, outside corporate security controls and endpoint monitoring, making this a practical gap in current SEG architecture.

JFrog Artifactory CVE-2026-82329 went from patch to active exploitation in three days. The critical authentication bypass in Artifactory's default configuration, rated CVSS 9.8, was patched on August 28, and by September 1 watchTowr had already observed attackers exploiting it to mint administrative tokens, enumerate users and credentials, and probe federated access relationships. The real danger is supply chain reach: Artifactory stores binaries, containers, packages, and AI models consumed by build and deployment pipelines, and administrative compromise lets attackers replace trusted artifacts, potentially injecting malicious code into downstream production systems. JFrog Cloud environments have been secured by the vendor, but self-hosted deployments must upgrade immediately.

Google shipped Chrome 152 patching 26 security vulnerabilities, two rated critical: CVE-2026-84353 in Shared Tab Groups and CVE-2026-84352 in WebGL, both use-after-free bugs. Eleven additional flaws received high-severity ratings, including use-after-free bugs in the Proxy, Browser, and Dawn components, an uninitialized resource issue in V8, and a GPU buffer overflow. The breadth of memory-safety issues across different Chrome subsystems in a single release, combined with Firefox 152 shipping critical patches in the same cycle, underscores the ongoing memory-corruption attack surface in modern browsers.

HPE Networking patched 52 vulnerabilities in Fabric Composer versions 7.3.3 and earlier, two of which scored CVSS 10.0. CVE-2026-76658 in the SSH daemon lets an unauthenticated remote attacker execute arbitrary commands as a privileged OS user, while CVE-2026-76657 grants unauthenticated admin access to the Fabric Composer interface. Both require no authentication and no user interaction. Fabric Composer manages data-center network fabrics, so full compromise gives an attacker the ability to reconfigure network infrastructure, intercept management traffic, and pivot into broader environments. HPE said it was not aware of active targeting at disclosure, but the pre-auth RCE at maximum severity on a network management platform makes exploitation a matter of when, not if.

An international law enforcement operation dismantled infrastructure belonging to the Sality P2P botnet, which had been active since at least 2003. The U.S. DOJ, FBI, and DCIS seized Sality-linked domains, while partners in Bulgaria, Hungary, and Romania took down European-hosted servers. CrowdStrike executed a P2P sinkhole isolating infected machines from control channels, and Europol confirmed the operation targeted a network linked to over 11 million infected IP addresses over its lifespan. At disruption, two active sub-networks were pushing EggJagger malware through clipjacking attacks, with approximately 15,000 active devices remaining infected. CrowdStrike tracks the operator as SALTY SPIDER, believed to be based in Russia's Republic of Bashkortostan.

CVE-2026-80047 in Hugging Face Transformers versions 4.49.0 through 5.8.1 allows attacker-controlled Python code to be written to disk before the user approves remote code execution. The flaw affects the load_custom_generate() workflow, which downloads the file and copies it into the local cache at ~/.cache/huggingface/modules before the trust_remote_code consent prompt is evaluated. CERT/CC documented this as VU#456290. The practical risk depends on downstream tooling: if any automated process or IDE reads from the Hugging Face module cache, the planted code could execute without the user ever seeing the trust prompt.

The Singularity Linux rootkit evades Elastic Defend by abusing the product's trusted_pids eBPF map. The rootkit loader dynamically identifies the relevant Elastic BPF program and map, then adds its own process ID to the trusted_pids hash. When the BPF program checks whether the loading process is trusted, it finds a match and exits before collecting module metadata or generating telemetry. The proof of concept was tested against Elastic Defend 9.5.2 on Ubuntu with kernel 6.8.0-138-generic. Because the rootkit dynamically locates the map at runtime rather than relying on a fixed map identifier, the evasion adapts across Elastic versions without hardcoding offsets. Organizations running Elastic's endpoint agent on Linux should monitor for unexpected eBPF map modifications.

The Gentlemen ransomware group deployed a previously undocumented command-and-control framework called TukTuk, discovered on a Finland-hosted Hetzner server containing the complete TukTuk v2.0 project archive with separate Windows and Linux agents, a backend server, and an operator control panel. The panel includes credential harvesting, screenshot capture, file management, and process control capabilities, with a dedicated credential display feature surfacing browser-saved passwords, RDP history, and WiFi profiles. Researchers linked the infrastructure through a GentleKiller driver hash match and a Greenshot DLL sideloading package consistent with known group techniques. The cross-platform nature of the framework and the breadth of defensive-evasion tooling suggest the group is investing heavily in operational infrastructure rather than relying on commodity RATs.

Watchlist: deploy SonicWall SMA1000 hotfixes before the September 5 CISA deadline and re-image affected appliances; patch Sangoma Switchvox to 8.4.0.2 or later and audit for reverse shell connections and unauthorized database modifications; emergency-patch Cisco Nexus 9000 switches with Silicon One before attackers weaponize the unauthenticated RCE; update All-in-One WP Migration to the fixed version and audit WordPress installations for signs of trackback-based injection; review external Teams communication policies and restrict or flag external messages from new contacts; disable or sandbox Git core.fsmonitor in AI coding agent environments until vendor patches ship; deploy StreamRAT indicators and audit Android devices for Accessibility abuse and VPN permission anomalies; audit Apache modules on Linux web servers for unauthorized reverse-proxy configurations; add QR-code decoding to email security pipelines; patch self-hosted JFrog Artifactory and audit for unauthorized admin tokens created since August 28; push Chrome 152 and Firefox 152 updates; patch HPE Fabric Composer; verify Hugging Face Transformers is patched and audit models loaded from external sources; monitor Elastic Defend environments for unexpected eBPF map modifications; track TukTuk indicators and review EDR telemetry for driver-based defense tampering.