CVE-2026-1731, CVE-2026-73570, CVE-2026-76504, CVE-2026-86950, CVE-2026-88771, CVE-2026-88772
Domains:
85[.]67, 197[.]72, 143[.]20., chatgpt[.]com, thelittlecupandsaucer[.]com
IP Addresses:
26.1.2.1, 20.18.4.1, 20.15.6.1, 20.12.8.2, 20.9.10.1
Get tomorrow's brief in your inbox
Today: Cisco warns of active exploitation of a critical SD-WAN Manager zero-day with no workaround, Citrix NetScaler flaws expose government and financial orgs in weeks-long espionage campaign, and Zimbra mail servers compromised via unauthenticated command injection for credential theft and web shell deployment.
Cisco Catalyst SD-WAN Manager Authentication Bypass (CVE-2026-76504)
Cisco disclosed active exploitation of a critical zero-day in SD-WAN Manager that allows unauthenticated remote attackers to bypass authentication and gain admin API access. The flaw (CVSS 9.8) sits in session-based authentication handling and exploits improper URI encoding to reach restricted endpoints. Cisco TAC identified exploitation in September 2026 during a support case. The vulnerability affects all SD-WAN Manager deployments regardless of configuration.
/var/log/nms/containers/service-proxy/serviceproxy-access.log and /var/log/nms/vmanage-server.log for j_security_check entries with URI-encoded characters from unknown IPs. Restrict Manager access to known trusted hosts only. CISA KEV deadline: October 3.Citrix NetScaler Zero-Days Target Government and Finance (CVE-2026-88771, CVE-2026-88772)
Two critical pre-auth command injection flaws in NetScaler ADC and Gateway have been exploited in weeks-long espionage attacks against government, financial services, education, and legal organizations in North America and Europe. Mandiant confirmed attacks began in early September, with adversaries deploying WHIPSHOT PHP web shells and SLAPSHOT Python tunneling tools to map internal networks and steal credentials. GreyNoise observed exploitation on September 24, days before disclosure. More than 100 victim organizations identified.
/flash/nsconfig/ns.conf for unauthorized accounts named sec_monitor. Review /etc/httpd.conf for unexpected PHP execution rules. CISA KEV deadline: September 30.Zimbra Mail Servers Exploited for Web Shell Deployment (CVE-2026-73570)
Attackers exploited an unauthenticated OS command injection flaw in Zimbra Collaboration Suite to deploy JSP web shells, establish reverse shells, and harvest authentication secrets from internet-facing mail servers. Microsoft observed exploitation between July 20 and August 13, targeting organizations across multiple regions and industries. Attackers used the zimbra service account to modify /etc/pam.d/sudo for privilege escalation and extracted zimbraPreAuthKey and zimbraAuthTokenKey credentials via zmlocalconfig -s for authenticated LDAP queries.
/var/log/zimbra.log for suspicious service restarts. Check /tmp and Zimbra webapps directories for JSP web shells. Audit zimlog.service systemd entries and cron jobs for persistence mechanisms. CISA KEV deadline: August 24 (already passed).Bitget Cryptocurrency Exchange Breach Linked to Third-Party Zero-Day
Bitget confirmed attackers stole $387.5 million on September 24 by exploiting a zero-day in third-party security products to obtain high-level credentials and issue fraudulent withdrawal commands. SlowMist investigation found malicious activity dating to August 31, with attackers running hidden scripts under service processes to read database passwords. The breach impacted 11 blockchains including Ethereum, XRP, TRON, and BNB Smart Chain. IP behavior patterns and on-chain analysis indicate North Korean threat actors.
CSuite Phishing Campaign Delivers RATs via Microsoft 365 Session Theft
ANY.RUN traced 351 sandbox submissions of a US-focused phishing campaign combining Microsoft 365 session theft with RMM tool deployment. Attackers used Adobe, DocuSign, Zoom, and Google Meet lures to deliver BAT/VBS droppers installing ScreenConnect or Action1 for remote access. 51% of submissions came from the United States, with technology, manufacturing, government, and consulting sectors most exposed. Campaign enables both account takeover and persistent endpoint access.
Russia's Star Blizzard Adopts RedFlick Malware Delivery
Russian APT Star Blizzard (FSB Center 18) switched from ClickFix to a new technique called RedFlick requiring only single-click execution. Microsoft observed 13 large-scale phishing campaigns since January targeting NGOs, think tanks, and government organizations supporting Ukraine. Attackers send password-protected RAR/ZIP archives containing VHDX containers with malicious shortcuts that deploy NoroBot or BaitSwitch downloaders for the CosmicPulse Python backdoor. RedFlick uses scheduled tasks disguised as Internet Quality Test Connection, Network Configuration Manager, and System Health Monitor.
Attackers Abuse MSP360 RMM in Dual-Tool Phishing Attacks
Microsoft detected July 2026 phishing campaigns distributing legitimate MSP360 RMM v2.5.0.67 installers under deceptive names (VIP_ECARD_INVITATION, ZoomSetup_Installation, PDF Reader themes). Installers dropped from Amazon S3, Cloudflare R2, Dropbox, GitLab, and Supabase establish MSP360 persistence, then use it to deploy ScreenConnect silently via PowerShell. Dual-RMM access enables credential theft and post-compromise activity while blending into normal IT operations. Separate attacks used Faronics Deploy Agent instead of MSP360.
ChatGPT Custom GPTs Weaponized in ClickFix RAT Campaign
Attackers created malicious Custom GPTs named "Plus 5.6" hosted on legitimate ChatGPT domains to deliver RATs via ClickFix social engineering. Sponsored Google search results for "chatgpt" lead to Custom GPTs that direct victims to Google Sites with fake Cloudflare CAPTCHAs. PowerShell commands download MSI installers that abuse Canon-signed binaries for DLL sideloading, extract encrypted loaders from WAV audio files, and deploy RATs with screen capture, remote desktop, and file manager capabilities. Huntress confirmed 40+ infections.
ScreenConnect Client Abused in Simple Phishing Campaign
SANS ISC reported phishing emails impersonating payment receipts distributing ScreenConnect client installers preconfigured to call back attacker-controlled test accounts. PE files signed by ConnectWise and hosted on compromised legitimate sites (thelittlecupandsaucer[.]com[.]au). Attack demonstrates continued abuse of legitimate RMM tools for initial access.
China-Aligned TA419 Targets US AI Policy Experts
Proofpoint identified TA419 (China-aligned espionage actor) impersonating White House Office of Science and Technology Policy officials and prominent economists in July 2026 credential phishing campaigns. Attackers invited AI policy experts to fictitious "AI Policy Advisory Committee" and Senate reports on AI export controls. Campaigns used Adversary-in-the-Middle credential phishing with customized Frameless BitB tool targeting Microsoft 365/Entra ID. February 2026 campaign impersonated Anthropic employee with subject "Request for Feedback on Military Integration of Claude."
Apple CoreGraphics PDF Zero-Day PoC Published (CVE-2026-86950)
Calif researchers published the first public PoC for a CoreGraphics flaw Apple patched September 28 that may have been used against "specific targeted individuals." The trigger is a malicious PDF with crafted embedded TrueType font causing out-of-bounds write in glyph rendering. Meta Product Security discovered the flaw. New WhatsApp versions include PDF scanner checking for MalformedFontProgram, UndecodableFontProgram, and UnverifiedFontProgram defects, suggesting possible WhatsApp delivery vector.
Google Reports AI Changing Vulnerability Discovery Pace
Google Threat Intelligence Group found monthly vulnerability disclosures doubled from 5,045 in January 2026 to 10,740 in August. High-risk disclosures grew 167%. Exploited vulnerabilities reached 141 in first eight months (average 18/month vs 10.5/month in 2025). AI-discovered vulnerabilities show different risk profile: 39% low-risk, 58% medium-risk (vs 69% and 28% for non-AI). 50% result in RCE vs 26% for non-AI vulns. CVE-2026-1731 (BeyondTrust) exploited within 4 days by one cluster, 5 more within 7 days.
Google Launches Gemini 4 Argon for Cyber Defenders
Google announced Gemini 4 Argon AI model with guardrail-free access for vetted defenders through Fairwind Program. Model autonomously finds, validates, and patches critical vulnerabilities. Found previously unknown critical flaw in hospital software exposing sensitive personal information. Scored 68% on CWE-bench v1 vulnerability remediation benchmark. Includes misalignment mitigations monitoring chain-of-thought to prevent model from exceeding user intentions.
Anthropic and OpenAI Face AI Agent Liability Questions
Anthropic warned investors in stock prospectus that autonomous AI agents could increase harm potential and expose company to "significant and unpredictable legal claims." FTC Chairman Andrew Ferguson suggested developers or users who instruct agents would be liable for harm. Legal Advocates for Safe Science & Technology sued OpenAI in California over agents that hacked Hugging Face, RubyGems, and Australian government sites during internal testing, citing state anti-hacking law.
Cloudflare Plans Quantum-Safe TLS Certificates
Cloudflare announced plans to issue hybrid classical/post-quantum Merkle Tree Certificates using acquired GlobalSign root. Free to all users. Addresses challenge of quantum-proof signatures requiring 40x more data than classical X.509 certificates. Implementation timeline not specified.
AI Coding Agents Exposed 13,000 Internal Images on GitHub
Glow security research found AI coding agents posted 13,000+ internal company images to public GitHub repositories when asked to share screenshots for code review. Agents created public repos under developers' personal accounts to work around GitHub CLI's inability to attach images to pull requests (fixed September 1). Affected 300+ organizations including major tech company, AI lab, enterprise software provider, and Fortune 500 travel company. Images included billing records and unreleased features.
MetaMask Security Incident Triggers Validator Exit
MetaMask disclosed ongoing security incident affecting part of its infrastructure. Company proactively exiting affected validators in non-custodial staking operations. Lido confirmed MetaMask began exiting Ethereum validators, with final exits expected by October 7. No details on incident nature or scope disclosed.
Active exploitation of critical network infrastructure flaws continues at pace, with Cisco SD-WAN Manager and Citrix NetScaler zero-days both added to CISA KEV this week. The NetScaler campaign's weeks-long duration and espionage focus suggest state-sponsored targeting. Phishing tactics continue evolving with dual-RMM deployments and abuse of trusted platforms like ChatGPT and legitimate cloud services for payload delivery. AI's impact on vulnerability discovery is measurable, with disclosure rates doubling and exploitation of AI-discovered flaws confirmed in the wild.