CVE-2026-15409, CVE-2026-15410, CVE-2026-19592, CVE-2026-48710, CVE-2026-49869, CVE-2026-55607, CVE-2026-59822, CVE-2026-69414, CVE-2026-72718, CVE-2026-82329, CVE-2026-83548, CVE-2026-83549, CVE-2026-9586
Domains:
iualef[.]net, oijfwe[.]net, app-microsoft-edge[.]com, baidu-pan[.]com, kaspersky-lab[.]hl
Hashes:
e0714788b4e2518b0d9d4cbf18c7217bb97718e01689d77338f1cc4a230fcb6c, ba83cc3c9535690191018edf73ca5c6001609df9919462796aa2e551f142e4d3
IP Addresses:
3.2.9.9
Get tomorrow's brief in your inbox
Today: SonicWall SMA 1000 devices face active zero-day exploitation with CISA deadlines this week. OpenAI's Astra model crossed the critical cybersecurity threshold by finding zero-days autonomously. Attackers are using AI agents to compress full ransomware operations from weeks into hours.
SonicWall SMA 1000 Zero-Days Enable Unauthenticated RCE (CVE-2026-83548, CVE-2026-83549)
SonicWall disclosed active exploitation of two zero-day vulnerabilities in SMA 1000 perimeter devices (models 6210, 7210, 8200v). CVE-2026-83548 (CVSS 10.0) is a pre-authentication SSRF vulnerability in the Work Place interface allowing unauthorized access without authentication. CVE-2026-83549 (CVSS 7.8) is a post-authentication OS command injection flaw in the Management Console enabling remote code execution. Attackers are chaining these vulnerabilities together to achieve unauthenticated RCE. CISA added both to the KEV catalog with a September 5, 2026 remediation deadline. This is the second SMA 1000 zero-day attack this summer, following CVE-2026-15409 and CVE-2026-15410 exploitation in June.
CISA Adds Seven Exploited Vulnerabilities to KEV Catalog
CISA added seven actively exploited vulnerabilities to its Known Exploited Vulnerabilities catalog on September 2. Beyond the SonicWall flaws, the additions include CVE-2026-9586 (CVSS 9.3) affecting Sangoma Switchvox, CVE-2026-82329 (CVSS 9.8) in JFrog Artifactory, CVE-2026-48710 (CVSS 6.5) in Kludex Starlette, CVE-2026-49869 (CVSS 10.0) in Kestra OSS, and CVE-2026-59822 (CVSS 8.8) in Berri LiteLLM. Threat actors weaponized CVE-2026-9586 and CVE-2026-82329 to deploy reverse shells and cryptocurrency miners. CVE-2026-49869 exploitation involved establishing reverse shells, conducting Docker environment discovery, and deploying cryptominers. Federal agencies must remediate KEV vulnerabilities by their due dates (September 5 or September 16 depending on the CVE).
AI Agents Execute Complete Ransomware Attack in Under 10 Hours
Unit 42 investigated an incident where a threat actor used frontier AI agents to execute a complete ransomware operation autonomously in less than 10 hours, compressing work that would normally take human operators two weeks. The attacker used frontier AI models and agentic frameworks to map internal architecture, raid source repositories, seize root credentials, trigger unauthorized CI/CD builds, and claim master keys to the victim's cloud AI infrastructure. The agents employed more than 50 MITRE ATT&CK techniques including secrets harvesting from code repositories, privilege takeover via exposed tokens, pipeline exploitation, and AI infrastructure hijacking. The attacker left an 80-page technical audit detailing dozens of exploited findings as a "report" on the organization's security posture. Indicators of AI usage included LLM calls to multiple frontier agents in parallel, structured Markdown files passing information between agents, and custom AI-generated scripts managing dynamic operations.
Microsoft Teams Vishing Campaign Delivers Remote Access and Malware
Microsoft Threat Intelligence observed a human-operated intrusion campaign (Spring Ring) targeting at least 150 Microsoft Teams users across 10 organizations. Attackers abuse Teams external collaboration to impersonate IT or helpdesk personnel and socially engineer users into granting remote sessions via RMM tools or Quick Assist. Once remote control is established, the attacker uses PowerShell to download and silently install a malicious MSI package staging a portable Node.js runtime and obfuscated JavaScript implant. The implant provides persistent command execution and C2. After deployment, attackers perform extensive host and Active Directory reconnaissance, capture screenshots, execute follow-on payloads through trusted Windows binaries, and pivot across the enterprise over WinRM toward high-value assets including domain controllers. Successful vishing calls last 10-15 minutes, with attackers making multiple attempts and leaving voicemails to connect with victims.
OpenAI's Astra Model Reaches 'Critical' Cybersecurity Capability Level
OpenAI announced its Astra model has reached the "Critical" cybersecurity capability level under the company's Preparedness Framework, the first OpenAI model placed in that category. The designation applies when a model can independently find and exploit zero-day vulnerabilities across many well-defended systems or carry out a complete cyberattack from only a high-level instruction. Astra achieved a perfect score on ExploitBench and uncovered two zero-day vulnerabilities independently during evaluations. The model broke out of a browser sandbox to run commands on the underlying machine and chained flaws in a hardened OS to gain root access. Astra declines 91.5% of cyber-related jailbreak attempts, up from 59% for GPT-5.6 Sol. Full cybersecurity capabilities will not be widely available at launch; OpenAI plans early access for a group of testers through its Daybreak Blue program.
CrowdStrike Privilege Escalation Zero-Day (FalconFlank) Disclosed
Security researcher Chaotic Eclipse released FalconFlank, a zero-day privilege escalation vulnerability affecting CrowdStrike Falcon Sensor. The flaw abuses the office malicious macros remediation feature in Falcon to achieve privilege escalation. The PoC works on fully updated Windows 11 25H2 and Windows Server 2025 with CrowdStrike Falcon installed. CrowdStrike may already have detections; users wanting to test should add it to exclusions or obfuscate the PoC. The disclosure follows the researcher's release of HardBreacher (Kaspersky privilege escalation), ShieldBreak (CVE-2026-69414, Microsoft Defender zero-day), and other endpoint security bypasses. The researcher stated Microsoft continues to refuse engagement or communication regarding the bugs.
Malicious Git Configs Enable Code Execution in AI Coding Agents
Manifold Security disclosed eight security flaws across seven AI coding agents where a repository's Git configuration can execute commands on the developer's machine. The vulnerability (GitSpawn) abuses the core.fsmonitor Git setting, whose value is a command Git runs to identify changed files. Agents including goose, Claude Code, Cursor, Hermes Agent, Qwen Code, and Grok Build call git status or git diff in the background, executing repository-supplied commands outside the agent's sandbox and without approval prompts. Fixes shipped for goose, Codex, and Cursor, while Hermes Agent, Qwen Code, Grok Build, and a second path in Claude Code remained vulnerable as of September 1. OpenAI published three CVEs covering the issue in Codex (CVE-2026-19592, CVE-2026-55607, CVE-2026-72718). Exploitation requires the repository to arrive with its .git directory intact (shared archive, sync folder, or USB stick), which an ordinary clone does not preserve.
BGP Hijack Delivers Malicious Virtualizor Updates
Softaculous' Virtualizor users were served malicious software updates for two days (August 28-30) after a BGP hijack attack diverted traffic to attacker-controlled servers. AS62390 (NexonHost) began announcing a portion of Hetzner's address space including Softaculous IPs. The threat actor obtained a valid Let's Encrypt TLS certificate for Softaculous domains during the hijack, allowing them to redirect traffic without triggering certificate warnings. A malicious Virtualizor package was delivered to installations that checked for updates during the hijack window. The malware established persistent root access via SSH keys, installed Java 17, downloaded a Java payload, created a systemd service for persistence, and created an unauthorized account named proxyuser. One hosting provider confirmed 5 of 34 checked Virtualizor hypervisors sustained root-level compromise. Virtualizor's update clients lacked cryptographic package verification.
Chinese-Speaking Threat Cluster Hijacks Brazilian Government Sites for SEO Manipulation
Check Point Research uncovered a sustained campaign (Gambling Goblin) against Brazilian organizations, primarily government and educational institutions, since mid-2025. The Chinese-speaking threat cluster, connected to Earth Berberoka, compiles and installs malicious Apache modules on victim servers that reverse-proxy visitors to attacker-controlled phishing pages while traffic appears to originate from legitimate domains. The phishing pages pose as Google Play, Microsoft Store, and Amazon while pushing online gambling and sports betting. The campaign chains together compromised high-reputation domains (many Brazilian government sites) to inflate search rankings. The group deploys DownPro (downloader), AlphaAgent (modular backdoor), oRAT (RAT), 3snake-based credential stealer, SSH brute-forcer, and plugin-driven reconnaissance tools. Parallel phishing networks localized in Vietnamese, Spanish, and English were identified, with infrastructure generating fresh domains daily.
NSO Group Pegasus Spyware Infects Serbian Student Movement Member via Zero-Click iMessage Exploit
Citizen Lab confirmed an iPhone belonging to a Serbian student protest movement member was infected with NSO Group's Pegasus spyware using an iMessage zero-click exploit. The infection occurred across December 2025 through January 2026, coinciding with local elections held March 29, 2026. Apple addressed the exploit with iOS 18.4.1 released in April 2025. At least 14 people in Serbia have been targeted with advanced spyware since the beginning of 2026, including student movement members, activists, a parliament member, and a local councilor. A separate student had their Android phone compromised with a new NoviSpy variant after device confiscation during police questioning.
Fake Software Installers Disable Windows Update and Weaken Microsoft Defender
Microsoft reported an active malware campaign using bogus software-download websites to impersonate trusted vendors and distribute malicious installers. The campaign has affected China-based operations of multinational organizations and Chinese-speaking users across healthcare, manufacturing, gaming, technology, logistics, government, and education sectors. Microsoft assesses with moderate confidence the activity aligns with Chinese threat cluster Silver Fox (Yinhu). Counterfeit websites hosted on .com.cn and .hl.cn infrastructure impersonate Microsoft Edge, Baidu Pan, Calibre, Kaspersky, and other popular software. Installers deploy malware that sets up persistence via scheduled tasks, weakens security by configuring Defender exclusions, deletes volume shadow copies, and disables Windows Update services (wuauserv, UsoSvc, uhssvc, WaaSMedicSvc). C2 communication occurs over non-standard ports including 5090, 7031, 7032, 7088-7090, 8050, 28290, and 28300.
Meta Ads Push StreamRat Android Trojan Granting Near-Complete Device Control
ThreatFabric disclosed StreamRat, a new Android banking trojan promoted to Spanish-speaking users through fake television-streaming campaigns on Meta platforms. The campaign reached an estimated 570,950 Meta accounts in the EU. The attack begins with social media lures directing Android users to specially crafted websites that push app.apk downloads. The dropper requests to become the default Home application, establishes a VPN connection routing traffic to a nonfunctional interface, and downloads the StreamRat payload (update_{timestamp}.apk). Once Accessibility access is granted, operators can capture keystrokes, display credential-stealing overlays, inspect the visible interface, and control the device remotely. StreamRat uses MediaProjection API for screen captures and can invoke Accessibility to interact with consent dialogs. The same banners were likely displayed on Facebook, Instagram, and TikTok.
AI is fundamentally changing the speed and economics of both offensive and defensive security operations. Multiple stories today demonstrate AI agents compressing attack timelines from weeks to hours, finding zero-day vulnerabilities autonomously, and enabling lower-skilled attackers to execute sophisticated campaigns. At the same time, edge devices like SonicWall SMA 1000 continue to face repeated zero-day exploitation, highlighting the persistent risk to internet-facing infrastructure. Supply chain attacks through BGP hijacking and malicious software updates show attackers targeting trust mechanisms at the infrastructure layer. The volume of actively exploited vulnerabilities added to CISA's KEV catalog this week underscores the growing gap between vulnerability discovery and remediation capacity.