The Exploit Bulletin

Archives
Log in
Subscribe
November 19, 2026

The Exploit Bulletin — Thursday, November 19, 2026: 10 issues require action

Thursday, November 19, 2026 — 10 issues require action. If you run none of the software below, you are done.

Affects: Cisco Expressway Series / TelePresence VCS Expressway · Vanderbilt University REDCap · ArangoDB · BerriAI LiteLLM proxy / Red Hat Ansible Automation Platform / Red Hat OpenShift AI · Apache Roller · Dagu · TopThink ThinkPHP · FIT2CLOUD MeterSphere · Elementor Pro · RViz


1. Static database-service cookies give unauthenticated root command execution on Cisco Expressway (CVE-2026-76466)

REMOTE CODE EXECUTION · CRITICAL · CVSS 9.8

VulnCheck added this flaw to its known-exploited catalog on 2026-11-18 citing an F6 report, and a public exploit is available in VulnCheck XDB; with Cisco offering no workaround, an unpatched Expressway-E on the internet edge is a root shell for anyone holding the static cookies.

The Expressway Series database service relies on persistent cookies and, when TLS Verify Mode is enabled for clustering, does not actually validate client certificates. An unauthenticated remote attacker with the cookies can connect to the database service and invoke RPC commands that execute arbitrary operating system commands as root on Expressway-C and Expressway-E devices.

Affected: Cisco Expressway Series (Expressway-C and Expressway-E) / TelePresence VCS Expressway X8.1.1, X8.2.2, X8.6, X8.8.3, X8.9, X8.11.0, X8.11.2, X8.11.3, X12.5.1, X12.5.2

How to Test: Check the running software version on every Expressway-C and Expressway-E against the affected list above. Review device logs for connections to the database service and RPC invocations from addresses that are not cluster peers, and audit for unexpected root-level processes, new accounts, or configuration changes; if TLS Verify Mode for clustering is enabled, assume client certificates were not being validated on affected builds.

How to Patch: Install the Cisco software updates that fix bug CSCww27732 (advisory cisco-sa-vcs-static-cred-hr7EWZ8t). Cisco states there are no workarounds; until the upgrade is complete, restrict network reachability of the database service to cluster peers only.

Evidence: VulnCheck KEV · VulnCheck XDB public exploit

Full entry with sources →


2. Unauthenticated code injection via public survey route in REDCap (CVE-2026-90817)

CISA KEV (due 2026-11-21) · REMOTE CODE EXECUTION · CRITICAL · CVSS 9.8

CISA added this unauthenticated RCE to KEV on 2026-11-18 with a three-day remediation deadline, and a REDCap server left on a vulnerable build with public surveys enabled can be taken over by anyone holding a survey link.

An unauthenticated request made in a public survey context can reach an unintended controller route and supply a crafted file-path or stream parameter to REDCap's data import processing, resulting in arbitrary code execution on the server. The only prerequisite is a valid public survey hash, which is shared openly by design when surveys are distributed.

Affected: Vanderbilt University REDCap 13.3.0

How to Test: Check the REDCap version shown in the Control Center and compare it with the latest consortium release. Review web server access logs for requests carrying a public survey hash that hit controller routes unrelated to survey rendering, particularly data-import handlers with file-path or stream parameters. Look for unexpected files in the REDCap web root, unusual outbound connections, and cryptominer processes such as XMRig, which the Palo Alto Unit 42 note cited by VulnCheck associates with this flaw.

How to Patch: Upgrade REDCap to the current release distributed through the REDCap Consortium community site (https://projectredcap.org/resources/community). Until the upgrade is complete, disable or restrict public surveys, place the instance behind an access gateway so public survey URLs are not reachable from the open internet, and regenerate public survey links afterwards.

Evidence: CISA KEV · VulnCheck KEV · CISA SSVC: active · CISA alert: added to KEV based on evidence of active exploitation · Public exploit in VulnCheck XDB

Full entry with sources →


3. URL-encoding authentication bypass in ArangoDB grants unauthenticated superuser access (CVE-2026-126755)

AUTHENTICATION BYPASS · CRITICAL · CVSS 9.3

VulnCheck listed this as exploited on 2026-11-18, and any reachable ArangoDB below 3.12.11 gives unauthenticated attackers superuser read, write and delete over every database, including password hashes.

ArangoDB makes its authentication decision on the raw request path but dispatches on the URL-decoded path, so encoding the leading underscore as %5f (e.g. /_db/_system/%5fapi/simple/remove-by-example) is treated as a public route, escalated to superuser, and routed to the real /_api/* action. On a default installation this allows unauthenticated read, write and delete of any document in any database, including the _users collection with password hashes.

Affected: ArangoDB 3.12.10.1 and earlier (fixed in 3.12.11)

How to Test: Check the running arangod version; anything at or below 3.12.10.1 is affected. Review HTTP access logs for request paths containing %5f or %5F after /_db//, such as /_db/_system/%5fapi/simple/first-example or /_db/_system/%5fapi/simple/remove-by-example, and audit the _users collection and other data for unexpected reads, modifications or deletions.

How to Patch: Upgrade ArangoDB to 3.12.11. Until then, block or reject requests whose path contains %5f/%5F at a reverse proxy in front of ArangoDB and restrict network access to the database port to trusted hosts.

Evidence: VulnCheck KEV · VulnCheck advisory (VulnCheck KEV 2026-11-18; Shadowserver exploitation since 2026-09-16)

Full entry with sources →


4. Code execution via bytecode rewriting at /guardrails/test_custom_code in LiteLLM proxy (CVE-2026-40217)

CISA KEV (due 2026-12-02) · REMOTE CODE EXECUTION · CRITICAL · CVSS 8.8

CISA added this flaw to KEV on 2026-11-18 while VulnCheck sensors were still logging exploitation attempts on 2026-11-19, so a LiteLLM proxy reachable with any valid or leaked API key hands an attacker the host and every upstream provider credential it stores.

The LiteLLM proxy's /guardrails/test_custom_code endpoint accepts caller-supplied code and executes it on the server through bytecode rewriting. Any caller holding a low-privilege API key can run arbitrary commands on the proxy host, which typically holds the organization's LLM provider keys and routes all model traffic.

Affected: BerriAI LiteLLM proxy (builds exposing /guardrails/test_custom_code); Red Hat Ansible Automation Platform 2.6 prior to RHSA-2026:24866; Red Hat OpenShift AI prior to 3.3.4 (RHSA-2026:30056)

How to Test: Confirm the installed LiteLLM proxy version and whether the /guardrails/test_custom_code route is served (request it with a valid key and check for a non-404 response). Review proxy and reverse-proxy access logs for POST requests to /guardrails/test_custom_code, especially from keys or source addresses that do not belong to guardrail administrators; VulnCheck sensors have seen attempts since 2026-07-30. Check for unexpected child processes spawned by the LiteLLM process, new outbound connections, and audit which API keys exist and who holds them.

How to Patch: Upgrade LiteLLM to the current release that removes the exploitable endpoint behaviour. Red Hat customers apply RHSA-2026:24866 for Ansible Automation Platform 2.6 and update OpenShift AI to 3.3.4 per RHSA-2026:30056. Until the upgrade is applied, block /guardrails/test_custom_code at the reverse proxy or firewall, restrict the proxy to trusted networks, and rotate LiteLLM API keys and upstream provider credentials if exposure is suspected.

Evidence: CISA KEV · VulnCheck KEV · CISA SSVC: active · CISA alert: added to KEV based on evidence of active exploitation · Public exploit in VulnCheck XDB

Full entry with sources →


5. Pre-authentication XML-RPC deserialization RCE in Apache Roller (CVE-2026-82384)

REMOTE CODE EXECUTION · CRITICAL · CVSS 9.8

Apache's advisory of 2026-11-17, reported on 2026-11-18, confirms attackers are exploiting this pre-authentication deserialization flaw and widens the affected range beyond 6.1.5, so Roller servers that assumed disabling XML-RPC was enough remain open to full takeover.

Roller's XML-RPC servlet accepts the library's vendor extension types, which are deserialized during request parsing before any authentication check, letting an unauthenticated attacker feed attacker-controlled bytes into Java deserialization and execute code. The servlet is mapped unconditionally, so requests are parsed even when the webservices.enableXmlRpc feature is disabled and no non-default configuration is needed.

Affected: Apache Roller 6.1.5 (Apache's follow-up advisory states additional earlier releases are also affected)

How to Test: Confirm the Roller version in the admin interface or deployed WAR; anything older than 6.1.6 should be treated as exposed regardless of the webservices.enableXmlRpc setting. Review servlet container and reverse-proxy access logs for POST requests to the XML-RPC endpoint, particularly with bodies containing non-standard extension value types, and treat any such traffic on an instance with XML-RPC disabled as hostile. Check for unexpected processes spawned by the Java servlet container, new files in the webapp directory, and unexplained outbound connections.

How to Patch: Upgrade to Apache Roller 6.1.6 or later, which disables the vendor extension types and gates the endpoint on webservices.enableXmlRpc. If the upgrade must wait, block the XML-RPC servlet path at the reverse proxy or firewall, since setting webservices.enableXmlRpc to disabled does not close the endpoint in affected builds.

Evidence: VulnCheck KEV · BleepingComputer quotes Apache's 2026-11-17 advisory stating the issue is being actively exploited · Public exploit in VulnCheck XDB

Full entry with sources →


6. Default no-auth mode in Dagu allows unauthenticated command execution via /api/v2/dag-runs (CVE-2026-126752)

REMOTE CODE EXECUTION · CRITICAL · CVSS 9.3

VulnCheck listed this as exploited on 2026-11-18, and any reachable Dagu instance running default settings hands OS command execution to anyone who sends one POST request.

Dagu ships with auth.mode set to none, so the POST /api/v2/dag-runs endpoint accepts an inline YAML DAG spec from anyone and immediately runs its shell commands on the host. Every default Docker or documented install without explicit authentication configuration gives an attacker command execution as the dagu process user.

Affected: Dagu >= 1.23.0, < 2.0.0 (default configuration with authentication disabled)

How to Test: Check the running Dagu version (e.g. the image tag of ghcr.io/dagu-org/dagu); versions from 1.23.0 up to but not including 2.0.0 are affected, and an instance is exposed if DAGU_AUTH_MODE is unset or set to none. Review access logs for POST requests to /api/v2/dag-runs containing a spec field from unexpected sources, check the DAG run history for unknown runs, and look for unexpected files such as /tmp/pwned or other commands executed as the dagu user.

How to Patch: Upgrade to Dagu 2.0.0, which enables authentication by default and adds a first-run admin setup. Until you can upgrade, set DAGU_AUTH_MODE to basic or builtin so the API requires credentials, and keep port 8080 off the internet.

Evidence: VulnCheck KEV · VulnCheck advisory (VulnCheck KEV 2026-11-18; Shadowserver exploitation since 2026-03-11)

Full entry with sources →


7. _method=__construct request injection gives unauthenticated command execution in ThinkPHP 5.0 (CVE-2019-25782)

REMOTE CODE EXECUTION · CRITICAL · CVSS 9.3

VulnCheck's canary sensors recorded exploitation attempts on 2026-11-18 and 2026-11-19 as it added the newly assigned CVE to its known-exploited catalog, and with public Exploit-DB code any internet-facing ThinkPHP 5.0 application below 5.0.24 can be taken over by a single POST request.

ThinkPHP 5.0.0 through 5.0.23 lets a POST body set _method=__construct, which re-invokes the Request constructor with attacker-controlled data and overwrites its filter property. Supplying filter[]=system together with a server[REQUEST_METHOD] value on routes such as ?s=captcha or ?s=index/index/index makes the framework execute arbitrary system commands without authentication.

Affected: TopThink ThinkPHP >= 5.0.0 < 5.0.24

How to Test: Confirm the framework version of each ThinkPHP application; in 5.0.24 library/think/Request.php only honors _method values of GET, POST, DELETE, PUT and PATCH, while earlier builds pass any value through. Search access logs for POST requests to ?s=captcha or ?s=index/index/index whose bodies contain _method=__construct, filter[]=system or server[REQUEST_METHOD]; treat matches as attempted or successful command execution and inspect the host for dropped webshells. The nuclei template thinkphp-5023-rce.yaml can verify exposure.

How to Patch: Upgrade to ThinkPHP 5.0.24 or later, which applies commit 4a4b5e64 to restrict the _method override to GET, POST, DELETE, PUT and PATCH. If an upgrade must wait, apply that commit's change to library/think/Request.php by hand, or block requests whose POST body contains _method=__construct at the reverse proxy.

Evidence: VulnCheck KEV · Exploit-DB 46150 · Packet Storm exploit

Full entry with sources →


8. Unauthenticated jar upload and invocation via /plugin endpoints gives RCE in MeterSphere (CVE-2022-51027)

REMOTE CODE EXECUTION · CRITICAL · CVSS 9.3

VulnCheck added this pre-authentication RCE to its known-exploited catalog on 2026-11-18 citing Shadowserver observations, and an unpatched MeterSphere instance reachable from the network can be fully taken over by anyone who uploads a jar.

MeterSphere exposed its /plugin/** endpoints to anonymous access, so an attacker can upload a malicious jar via POST /plugin/add and execute it through POST /plugin/customMethod, whose 'entry' and 'request' parameters feed Class.forName and method invocation. The result is remote code execution as the MeterSphere service without any credentials.

Affected: FIT2CLOUD MeterSphere 1.13.0 through 1.14.6; FIT2CLOUD MeterSphere 1.15.0 through 1.16.3

How to Test: Check the installed MeterSphere version: anything in the 1.13.0–1.14.6 or 1.15.0–1.16.3 ranges is vulnerable. Review reverse-proxy and application access logs for unauthenticated POST requests to /plugin/add and /plugin/customMethod, and inspect the server for unexpected plugin jar files or child processes spawned by the MeterSphere Java process.

How to Patch: Upgrade to MeterSphere 1.14.7 or later on the 1.13/1.14 line, or 1.16.4 or later on the 1.15/1.16 line; the fix (commit 07155cf833) removes the anonymous filter-chain rule for /plugin/ in ShiroUtils. If an upgrade must wait, block external access to /plugin/ at the reverse proxy.

Evidence: VulnCheck KEV · VulnCheck advisory cited by VulnCheck KEV as exploitation source

Full entry with sources →


9. Unauthenticated arbitrary file upload to RCE in Drag and Drop File Upload for Elementor Forms (CVE-2026-18351)

PUBLIC EXPLOIT · NO ATTACKS CONFIRMED · REMOTE CODE EXECUTION · CRITICAL · CVSS 9.8

A Nuclei template and two VulnCheck XDB exploits for this unauthenticated upload are public and VulnCheck added the flaw to its known-exploited catalog on 2026-11-17; a site still running 1.6.0 or earlier can be handed a PHP web shell with a single crafted POST.

The plugin's elementor_file_upload function validates uploads in is_file_type_valid() using the attacker-controlled 'type' parameter as regex keys into its MIME allowlist, and sanitize_file_name() later normalizes a crafted extension to a PHP extension. An unauthenticated visitor can therefore upload an executable PHP file to the site and run code on the server.

Affected: Drag and Drop File Upload for Elementor Forms (WordPress plugin) <= 1.6.0

How to Test: In WordPress admin under Plugins, check the installed version of Drag and Drop File Upload for Elementor Forms; 1.6.0 and earlier are affected. Review web server access logs for unauthenticated POST requests hitting the elementor_file_upload action with unusual 'type' values, and inspect the plugin's upload directory for .php files or files with crafted multi-part extensions — any PHP file in an upload location is a compromise indicator.

How to Patch: Update the plugin through the WordPress plugin updater to the release that follows 1.6.0 (the fix is changeset 3628051). If you cannot update immediately, deactivate the plugin until you can.

Evidence: VulnCheck KEV · Merged Nuclei template · VulnCheck XDB public exploit · VulnCheck XDB public exploit (second entry)

Full entry with sources →


10. Unauthenticated heap overflow in mesh handling for PointCloud and Marker displays allows network-adjacent code execution (Open Robotics RViz)

UNVERIFIED PUBLIC REPORT · NO CVE · REMOTE CODE EXECUTION · HIGH

ZDI published this as a 0-day on 2026-11-18 after Open Robotics did not respond since the July report; details are now public with no fix.

ZDI reports a heap-based buffer overflow in RViz's fillIndexBuffer handling of mesh resources used by PointCloud and Marker displays: the length of user-supplied data is not validated before it is copied to a heap buffer. An attacker on the same network as the ROS/DDS domain who can publish a crafted message can execute code in the context of the RViz user without authentication (ZDI CVSS 8.8, AV:A). ZDI reported it to Open Robotics on 2026-07-15 and received no response; there is no CVE, no vendor statement, no patch, and the advisory does not state which RViz releases are affected, so all versions should be treated as affected.

Affected: not stated — the report concerns Open Robotics RViz

How to Test: Check whether RViz or rviz2 is installed (dpkg -l | grep rviz, or ros2 pkg list | grep rviz) and whether operators run it on hosts joined to a ROS/DDS domain reachable by untrusted machines; check if any Marker or PointCloud displays load mesh_resource data from topics others can publish to. Indicators: RViz crashes when subscribing to marker or point-cloud topics, or marker messages with mesh resources arriving from unexpected publishers (ros2 topic info -v).

How to Mitigate: There is no fix. Restrict the DDS/ROS domain to a trusted, segmented network and firewall DDS discovery and data ports from untrusted hosts; enable SROS2 with DDS security so only authenticated, authorized nodes can publish to the topics RViz displays; run RViz under a low-privilege account and avoid running it on operator workstations exposed to shared or field networks until Open Robotics ships a patch.

Evidence: ZDI-26-1027 0-day advisory

Full entry with sources →


Read on the web · Every past edition

The Exploit Bulletin is free and daily. It publishes only what security teams must act on today — nothing else. Forward it freely.

Spot an error, or an exploit we missed? Reply here or email [email protected].

Don't miss what's next. Subscribe to The Exploit Bulletin:
← Newer The Exploit Bulletin — Friday, November 20, 2026: 4 issues require action Older → The Exploit Bulletin — Wednesday, November 18, 2026: 6 issues require action
www.exploitbulletin.com
jbac.co
LinkedIn
Powered by Buttondown, the easiest way to start and grow your newsletter.