Skip to content

Response, New York. Noah Park: Check Point Patches Actively Exploited Management Server Path Traversal. Detection, Dallas. Ava Okello: Report Finds Nearly Half of Deployed Detections Need Attention Before SOCs Can Trust Them.

SOCtember

Always First. Fast SOC News.

Rapid7 screenshot of a Metasploit console running the beyondtrust_rs_mrs_file_upload module against a BeyondTrust Remote Support test appliance, showing the check that the target appears vulnerable, the PHP upload to the public directory, the request to the extensionless PHP endpoint and an opened Meterpreter session with sysinfo output from the appliance.
Image: Rapid7 (fair use). Figure 1 from Rapid7's Oct. 30, 2026, analysis, exploiting the anonymous file-transfer service with Metasploit in a lab.

Response

Rapid7 Releases a Metasploit Module for the BeyondTrust Remote Support Flaw China-Nexus Hackers Exploited as a Zero-Day, and Shows the Log Lines That Can Reveal an Attack

CVE-2026-102764 lets an anonymous support-session customer forge a message the appliance treats as local, write a PHP file to the public directory and run code as the site user. Rapid7 says a hunt that only looks for requests ending in .php would miss its successful test, and that per-session logs are cleared when a worker exits.

Noah Park, ResponseNew York6 min read

NEW YORK - Rapid7 on Friday published a full technical analysis and a working Metasploit module for CVE-2026-102764, the BeyondTrust Remote Support flaw that Mandiant says a China-nexus group exploited as a zero-day, giving any attacker a ready-made way to write files to an unpatched appliance and run code on it without a password.

The analysis, by Rapid7's Jack Heysel, shows that the entry point is the public customer connection that starts a support session. "No account credentials or user interaction are required," he wrote. The module, exploit/linux/http/beyondtrust_rs_mrs_file_upload, uploads a self-deleting PHP file to the site's public directory and then requests it to execute a payload.

BeyondTrust disclosed the flaw on Oct. 15 in advisory BT26-05, rated 9.3 under CVSS v4, describing it as improper origin validation in an internal messaging component. The company said the patch was applied to all Remote Support SaaS customers as of Oct. 9. Self-hosted appliances on versions up to 26.2.3 need the BT26-05-RS patch, which BeyondTrust lists for versions 24.3.1 through 26.2.3, and customers on releases older than 24.3.1 must upgrade before they can apply it. Rapid7 ran its tests on version 24.1.2, one of those older releases, and said other affected releases and fixed builds "have not been established" by its own testing.

A forged router ID that the appliance trusts

Rapid7 screenshot of the section titled A network message can claim to be local, with a table of the 36-byte message header fields, including the source router field changed to the reserved value 1, followed by decompiled code from the brain router's ingress check.
Image: Rapid7 (fair use).The message header fields the exploit manipulates, from the technical analysis.

The flaw sits between the appliance's internal message router, called brain, and a worker process, thin-scc, that the appliance starts for an anonymous customer's thin chat session. Rapid7 obtained two customer keys from the public POST /api/start_session endpoint, which needs no login, then connected through the /nw WebSocket endpoint.

Each internal message carries a source and destination address in a 36-byte header. By putting the reserved router value 1 in the source field, the attacker makes the worker accept the message as if it came from a local component. That opens an internal file-transfer service, including its native filesystem provider, and lets the attacker choose both the contents and the destination path of an uploaded file. Rapid7 used the same trick to read /proc/self/environ and find the installation path before dropping the payload. On the test appliance, the code ran as the site user, msfuser, UID 1000.

The researchers said the test appliance had public issue-menu session starts and thin chat enabled, which is the default.

What the attack leaves in the logs

Rapid7 screenshot of the section titled Requests to an uploaded PHP endpoint, showing access log lines with two POST requests to /api/start_session followed by POST and GET requests to a random extensionless path that return 200 and then 404.
Image: Rapid7 (fair use).Lab access log lines from the Metasploit module run.

Rapid7 published log excerpts from its lab runs on Oct. 28 and 29, and was careful about what each one proves. The site logs live under the installation's log directory: httpd_access.log, thin-scc.log, thin-scc-errors.log and brain.log.

In the access log, the module's run looked like two POST requests to /api/start_session within the same second, then a POST to a new, random, extensionless path that returned 200, followed by 404 responses for the same path after the file deleted itself. Session creation alone is normal Remote Support activity, Rapid7 noted, and a 200 followed by a 404 is not enough on its own.

Rapid7 screenshot of the section titled The URL may not contain .php, showing a request ending in .php returning 404 while the same path without the extension returns 200, followed by guidance on files and connections.
Image: Rapid7 (fair use).The extensionless request detail from the IOC section.

One detail matters for detection engineering. In an earlier test, a request for the uploaded file with a .php extension failed while the extensionless request ran it. "Detection that only searches for requests ending in .php would miss the successful request in this example," Rapid7 wrote.

Other traces Rapid7 suggested looking for:

  • Unexpected PHP files under the site's public directory, and received files ending in .php.n$t, which an interrupted transfer can leave behind even if the final URL returns 404. The suffix is also used by normal transfers, so it needs context.
  • In thin-scc.log, entries tagged with the LocalFileTransfer process showing a stream connecting to an internal address, followed by "nodeID not found" and "failed to create a mailbox for sending to: 0:0:0" errors, especially near unfamiliar PHP requests or outbound connections. Rapid7 cautioned that a different exploit implementation may leave different traces.
  • brain.log customer connection records, to tie activity to a client IP for the timeline.

Rapid7 also flagged a gap. While a worker is running, its logs sit under a per-session directory, data/tmp/sdcust_thin_client, and the wrapper appends them to the main log directory only when the worker exits, then clears the copies. Investigators should collect both locations during a live session. Debug output was disabled in the captured configuration files, and callback names such as UploadFileStarted did not appear in the collected logs, so they "should not be presented as default audit events," the researchers wrote.

What to do now

For self-hosted Remote Support appliances, confirm that the BT26-05-RS patch is installed or that automatic updates are enabled in the /appliance interface, and upgrade any instance older than 24.3.1 first. With a public module now available, Mandiant's earlier guidance carries more weight: treat any internet-reachable appliance that ran a vulnerable version during the exposure window as potentially compromised until reviewed, and use Rapid7's log locations to scope that review.

About the Author

Noah Park

Noah Park covers incident response and blue-team operations for SOCtember from New York.

Related stories

Response desk