exfil.tv AI-hosted Watch live
Vulnerability Analysis

Exploits hit Ivanti and Swann; Adobe patches roll

Vuln roundup: exploits and patches ·

CVE-2026-100133CVE-2026-103849CVE-2026-104935CVE-2026-92909CVE-2026-97229Adobe CommerceGitLabIvanti Connect Secure

In this episode

The facts

Sources

The panel

Jordan Pike AI host
CISA says attackers are exploiting CVE-2026-104935, a flaw in Ivanti Connect Secure, and has added it to its Known Exploited Vulnerabilities catalog.

Dex Moreno AI analyst
An internet-facing VPN like Connect Secure is a prime entry point. Once a malicious request lands, the attacker can execute code on the appliance, harvest admin credentials and pivot to internal services that trust the VPN for access. From there they could reach file servers, databases or SaaS portals, giving them a foothold deep in the network.

Dex Moreno AI analyst
Looks like the attackers skipped the doorbell and just walked in.

Jordan Pike AI host
One sentence.

Sam Kowalski AI analyst
Forensics should start with the VPN logs—look for unusual remote IPs, repeated failed logins and any successful auth that aligns with the exploit window. Capture the appliance image, hunt for new admin accounts, and isolate the device while you verify the vendor’s fix.

Jordan Pike AI host
Rapid7 reports attackers are exploiting CVE-2026-97229, a flaw in GitLab.

Dex Moreno AI analyst
GitLab’s web interface is exposed to the internet, so a remote attacker can send crafted requests that trigger the vulnerability, potentially achieving code execution on the GitLab server. That gives them a web-shell, which can be used to harvest repository secrets and token files, then hop to downstream services that pull code from the instance.

Sam Kowalski AI analyst
Pull the GitLab access logs and the audit events for any new keys or token creations. Check the logs for unexpected commands, and sandbox the server until the vendor’s remediation is applied.

Jordan Pike AI host
CISA says attackers are exploiting Swann Communications SWIFI-SE2KIC firmware, and a fix is available.

Dex Moreno AI analyst
The SWIFI-SE2KIC is a camera that talks to the cloud over HTTP. An attacker who reaches the device—often via default credentials or exposed ports—can run the malicious firmware, turning the camera into a foothold inside the network. From there they can sniff traffic, pivot to other devices, or use it as a beachhead for ransomware.

Sam Kowalski AI analyst
Review the camera’s access logs, look for unknown addresses or firmware version changes, and block outbound traffic until the patch is applied. A network-segmentation rule separating cameras from critical systems buys you time.

Jordan Pike AI host
CVE-2026-103849 is a flaw in Adobe Commerce and is classed as template injection (CWE-1336).

Dex Moreno AI analyst
Template injection lets an attacker inject malicious directives into the rendering engine, which can lead to remote code execution on the web server. If the Commerce instance is public, the attacker can craft a URL that executes on any visitor, or abuse admin pages to gain full control of the e-commerce backend.

Sam Kowalski AI analyst
Search web-server logs for query strings that include template syntax and watch for any unexpected admin activity after the vulnerability window.

Jordan Pike AI host
CVE-2026-92909 is a flaw in Adobe Commerce and is classed as incorrect authorization (CWE-863).

Dex Moreno AI analyst
Incorrect authorization means the application may let a low-privilege user perform actions reserved for admins, such as modifying product data or extracting customer records. An attacker can chain this with the earlier template issue to elevate privileges and exfiltrate data.

Sam Kowalski AI analyst
Audit role assignments in the Commerce admin panel, flag users with unexpected privileges, and keep an eye on logs for any signs of privilege escalation.

Jordan Pike AI host
CVE-2026-100133 is a flaw in Adobe Commerce and is classed as cross-site scripting (CWE-79).

Dex Moreno AI analyst
A reflected XSS vector lets an attacker embed malicious code in URLs that are served to victims. When a shopper clicks the crafted link, the script runs in their browser, stealing session cookies or performing actions on their behalf, effectively hijacking user accounts.

Sam Kowalski AI analyst
Enable a web-application firewall rule to block script tags in query parameters, scan the codebase for unescaped output, and monitor session logs for anomalous behavior.