🕵️ RESEARCH & DEEP DIVES
InfraTrust report warns attackers increasingly target network management systems
Eclypsium's September InfraTrust Pulse report finds attackers are increasingly exploiting critical vulnerabilities in network management and administrative platforms.- Applies to enterprises running infrastructure management consoles: Cisco FMC and ISE, SonicWall SMA 1000, Check Point, HPE Fabric Composer, EdgeConnect SD-WAN Orchestrator, NVIDIA Unified Fabric Manager, Dell SmartFabric Manager, SonicWall NSM On-Prem, and Arista management interfaces.
- InfraTrust tracked 158 advisories from 17 vendors (Aug 25–Sep 17) covering 1,699 vulnerabilities; 42 were critical, 71 remotely exploitable unauthenticated, and five reached CISA's KEV catalog.
- Cisco FMC flaws CVE-2026-20079 (CVSS 10.0 auth bypass allowing root command execution via crafted HTTP requests) and CVE-2026-20316 were chained in attacks, with the /var/tmp/license.tmp IOC observed; Cisco Talos linked activity to clusters UAT-12197, UAT-11823, and UAT-11988, including state-sponsored actors and ransomware gangs.
- Attackers used built-in FMC tools for reconnaissance, deployed tunneling utilities, harvested credentials, and in some cases deployed Qilin ransomware; a Linux implant 'timezone_check' on compromised FMC appliances was identified as a Cyclops Blink variant associated with Sandworm.
- SonicWall SMA 1000 flaws CVE-2026-83548 (SSRF) and CVE-2026-83549 (command injection) were chained for unauthenticated RCE; Cisco ISE CVE-2026-76460 and Check Point CVE-2026-85102, CVE-2026-85103, and CVE-2026-91843 are also critical, unauthenticated, and exploited or at imminent risk.
Attackers Poison ChatGPT, Gemini, and Google AI Overviews in Mass Disinformation, Phishing Campaign
Attackers are poisoning AI chatbot answers with fake support numbers and phishing links in a campaign dubbed Dark Sourcery.- Targets users of OpenAI ChatGPT, Google Gemini, and Google AI Overview; at least 374 companies affected, including Delta, Lufthansa, Qatar Airways, Chase, Bank of America, Airbnb, and TripAdvisor
- Tens of thousands of malicious pages contain fake support phone numbers, email addresses, login pages, and software-update details
- Attackers use SEO and content-distribution techniques, leveraging high-authority domains like universities and government sites, to get chatbots to retrieve and serve malicious content
- Unlike prompt injection, no explicit instructions are given to the AI; the poisoned content becomes part of the answer itself, bypassing AI defenses
- Researchers observed victims scammed into giving up payment card details via fraudulent phone numbers served in chatbot answers
GitLab 'Email work item' address leak lets attackers push code to private repos
Aikido Security researchers found that GitLab's 'Email work item to this project' address can be abused to push attacker-controlled code into private repositories.- Affects GitLab projects using the 'Email work item to this project' feature; the address embeds a long-lived, non-expiring glimt- incoming-email token that must stay secret.
- Addresses for different projects share the same account-level token; anyone holding it can create issues and merge requests as the token owner.
- Attackers swap the -issue suffix for -merge-request, attach a Git patch, and name a branch in the subject; GitLab applies the patch using the victim's permissions.
- A patch modifying .gitlab-ci.yml can trigger attacker-controlled CI/CD, exposing source code, CI/CD variables, job tokens, or secrets; a Maintainer's address can commit to protected branches like main.
- Incoming email bypasses IP allowlist restrictions, and sender spoofing is unnecessary since GitLab doesn't require verified sender addresses; GitLab treats the behavior as designed but updated docs and UI.
- IOC: exposed addresses containing the glimt- token prefix.
Sources 📄 Source: aikido.dev · 📎 Coverage: cybersecuritynews.com · 👁 via Cyber Security News, cryptika.com (discovered)
📋 ADVISORIES
UPDATE: OpenAI and Ukraine partner on 'Daybreak' program to defend power grids and water systems
OpenAI and Ukraine have agreed to a partnership providing AI cybersecurity tools and over $1 billion in subsidized tokens to protect critical infrastructure.- Applies to Ukraine's critical infrastructure operators, including electricity and water systems, facing roughly 6,000 cyberattacks per year amid the war with Russia.
- OpenAI's Daybreak program provides Ukrainian cybersecurity officials access to advanced AI models designed for cybersecurity work, plus over $1 billion in subsidized tokens.
- Planned uses include incident response, threat triaging, login analysis, system inventorying, code analysis, and vulnerability validation at machine speed, complementing rather than replacing human defenders.
- OpenAI says it feels urgency to strike similar agreements with other governments and industries; CEO Sam Altman met with U.S. power companies in July about AI grid defense.
- U.S. Senate Foreign Relations Committee staff cited Ukrainian infrastructure resilience as a model for broader adoption of defensive AI tools by U.S. operators.
📄 Source for Dell SCG Policy Manager flaw (CVE-2026-73591) exposes sensitive information in source code — dell.com
🔓 CVEs & KEV
CVE-2026-96275 — Red Hat Enterprise Linux 10 — CVSS 8.8 — Flatpak: flatpak: arbitrary write access as root via extra-data extractionA m...
CVE-2026-94384 — Amazon amazon-connect-salesforce-lambda — CVSS 6.4 — AWS Lambda Flaw Lets Attackers Bypass IAM Permissions and Access Cloud Services