This week, the team developed an exploit for CVE-2026-85706, a CVSS 10.0 unauthenticated arbitrary file read in GitLab CE and EE. CISA KEV and VulnCheck KEV both added it on September 11, 2026, a day before the CVE was even published to NVD, with a three-day federal remediation deadline under BOD 26-04. watchTowr reported honeypot probes roughly 20 hours after the patch shipped, and VulnCheck is already tracking more than a dozen scanners and public PoCs. GitLab instances hold source code, CI/CD secrets, and deploy tokens, so an unauthenticated read primitive is a valuable attack target.
Every public source, including the vendor advisory and the Nuclei template, states that exploitation requires at least one publicly readable project on the target. VulnCheck's testing showed that holds for only one of the two affected API routes: the other needs no public project and no valid project ID at all, so the exposed population is larger than the advisory implies. VulnCheck Target Intelligence finds 37K+ GitLab servers exposed to the internet, more than 7,600 of which are running an affected version.
GitLab reports no version number to an unauthenticated caller, which has always made GitLab exposure hard to scope from the outside. Alongside this exploit we shipped a build-hash fingerprinter that resolves an unauthenticated GitLab instance to a release, covering both CE and EE from GitLab 11.0 in 2018 through 19.3.2 today. Our exploit also ships with a Docker target, PCAPs, ASM queries, and Suricata, Snort, Sigma, and YARA rules.
The team added an exploit for CVE-2026-40058, a TOCTOU vulnerability in CrowdStrike's Falcon Sensor that allowed for elevation of privilege on Windows systems. CrowdStrike is a hugely popular EDR / XDR solution on enterprise networks, and the vulnerability received plenty of public attention when it was originally released as an unpatched zero-day. The team's exploit ships with a version scanner, Sigma rule, and a custom DLL proxy payload to ensure exploit execution stability.
The team added an exploit for a vulnerability in versions below 2026.3.1.14 of N-able N-central that yields unauthenticated remote code execution. CVE-2026-86218 is just one of a set of vulnerabilities recently disclosed for the RMM software and reported exploited by unattributed threat actors. It was added to the VulnCheck KEV list on September 6, 2026, and to CISA KEV two days later. VulnCheck Target Intelligence identifies a small number (under 20) of N-central instances vulnerable to CVE-2026-86218 at time of writing. Our exploit comes with a version scanner, network signatures, a YARA rule, PCAPs, and ASM queries.
The team developed an exploit for CVE-2026-86124, a critical unauthenticated RCE vulnerability in HKUDS AutoAgent, an academic LLM agent framework. VulnCheck's CNA published the CVE on September 4, 2026, and VulnCheck Canaries began detecting in-the-wild exploitation on September 18. The project appears to be abandoned with no fix planned, so any exposed instance remains exploitable indefinitely. We covered this issue as an example of a vulnerability category we expect to grow as LLM agent frameworks proliferate: internal "sandbox" control channels published to the public network with no authentication, where the sandbox process itself runs as root. The framework is primarily used as an internal developer tool and thus does not have a notable internet-facing footprint. Our exploit ships with a Docker target, a PCAP, and Suricata and Snort rules.
The team developed an exploit for CVE-2026-77539, a command injection vulnerability in Ubiquiti UniFi OS Server. UniFi OS Server is Ubiquiti's software deployment of the UniFi OS management platform, accessible through a browser-based UI and REST API. The sshHashedPassword field in the affected endpoint implements a weak sanitizer that can be bypassed, allowing an attacker to pass arbitrary commands that are subsequently executed by a bash script as root. VulnCheck Target Intelligence finds several thousand Ubiquiti UniFi OS Servers on the public internet. Our exploit comes with PCAPs, Snort and Suricata rules, and a Docker target.
The team also developed an exploit for CVE-2026-19295, a critical command injection in Langflow. IBM (the CNA) says the vulnerability is authenticated, but Langflow's default LANGFLOW_AUTO_LOGIN setting gives any visitor a free session, so our exploit runs unauthenticated end to end. No in-the-wild exploitation has been reported yet, but Langflow has been a rising attack target as AI platform adoption grows. VulnCheck Target Intelligence finds about 330 internet-exposed Langflow instances, the majority (250+) of which look to be vulnerable. Our exploit ships with a version scanner, a Docker target, PCAPs, and Suricata and Snort rules.
The team added an exploit for CVE-2026-9198, a critical unauthenticated RCE in Langflow that was added to VulnCheck KEV on August 4, 2026. Similar to CVE-2026-19295, Langflow's default configuration delivers any network caller a superuser session with zero credentials, so this is trivial pre-auth compromise of a platform that runs AI agent workflows and holds their credentials. VulnCheck Target Intelligence counts 218 exposed, vulnerable Langflow instances. Our exploit ships with a version scanner, a Docker target, and PCAPs.
Finally, completing this week's Langflow coverage, the team added an exploit for CVE-2026-55255, a high-severity vuln in Langflow's IDOR implementation that lets any authenticated user run another user's flow by swapping in that flow's ID. GitHub's advisory rates the attack complexity high, but our exploit reaches the victim's flow with a single authenticated API call once a known flow is provided. The vulnerability was added to both VulnCheck KEV and CISA KEV on July 7, 2026, not two weeks after it was first published. VulnCheck Target Intelligence counts 198 exposed, vulnerable Langflow instances. Our exploit ships with a version scanner, a Docker target, and PCAPs.
The team added a version scanner for CVE-2025-25249, a heap-based buffer overflow in the CAPWAP access controller daemon used by Fortinet FortiOS and FortiSwitchManager. Despite this vulnerability being a year old, the team researched this vulnerability by customer request following a recent report of active exploitation targeting internet-facing FortiGates. Per SOCRadar, a Russian-speaking, financially motivated operator has been exploiting the vulnerability since at least July 2026, probing tens of thousands of addresses with confirmed data exfiltration in at least two intrusions. The vulnerability was added to VulnCheck KEV on September 8, 2026 and CISA KEV a day later. VulnCheck Target Intelligence finds over 240K FortiOS instances on the internet.
The team also added exploits and detections for the following vulnerabilities to improve breadth and depth of historical coverage: