Sources: National Vulnerability Database, CVE.org, CISA KEV, FIRST EPSS
CVE-2026-125368 Overview
OmniAgent through commit 12b98c5 contains an approval bypass vulnerability in ReflexionAgent._is_safe_bash_command that treats any command starting with find as safe without inspecting -execdir arguments. Prompt-injected agents can run find with -execdir to invoke binaries like dd, nc, perl or openssl through an unsandboxed shell, overwriting files or executing code without human approval.
Critical Impact
Affected Products
- OmniAgent: see the vendor advisory for the exact affected versions
Discovery Timeline
- 2026-11-10: CVE-2026-125368 published to the National Vulnerability Database by disclosure@vulncheck.com
- 2026-11-10: last updated in the NVD database
Technical Details for CVE-2026-125368
Vulnerability Analysis
CVE-2026-125368 is a Security Vulnerability weakness (CWE-184) in OmniAgent. Based on its CVSS vector, the flaw can be reached over the network, needs no login, and relies on a user taking an action.
With a CVSS base score of 7.7, apply the vendor fix and confirm exposure on internet facing and business critical assets first.
Root Cause
The vendor description states: OmniAgent through commit 12b98c5 contains an approval bypass vulnerability in ReflexionAgent._is_safe_bash_command that treats any command starting with find as safe without inspecting -execdir arguments. Prompt-injected agents can run find with -execdir to invoke binaries like dd, nc, perl or openssl through an unsandboxed shell, overwriting files or executing code without human approval.
Attack Vector
Based on the CVSS vector, the flaw can be reached over the network, needs no login, and relies on a user taking an action. Attack complexity is high complexity.
No exploitation in the wild is confirmed by CISA at the time of the last update. Check the references below for any vendor or researcher proof of concept.
CVSS:4.0/AV:N/AC:H/AT:P/PR:N/UI:P/VC:H/VI:H/VA:H/SC:N/SI:N/SA:N/E:X/CR:X/IR:X/AR:X/MAV:X/MAC:X/MAT:X/MPR:X/MUI:X/MVC:X/MVI:X/MVA:X/MSC:X/MSI:X/MSA:X/S:X/AU:X/R:X/V:X/RE:X/U:X
Detection Methods for CVE-2026-125368
Detection Strategies
- Inventory every instance of OmniAgent and compare the running version with the affected range above.
- Check external attack surface scans for internet facing instances of the component.
- Run an authenticated vulnerability scan once your scanner ships a check for this CVE.
Monitoring Recommendations
- Forward OmniAgent application, access and audit logs to your SIEM for retrospective hunting.
- Alert on new processes, configuration changes or outbound connections from the affected service.
- Track the vendor advisory and the CISA KEV catalog for changes in exploitation status.
How to Mitigate CVE-2026-125368
Immediate Actions Required
- Apply the fixed version of OmniAgent named in the vendor advisory.
- Prioritize internet facing and business critical instances first.
- Rescan after the change to confirm the exposure is closed.
Patch Information
Use the fixed version listed in the vendor advisory for OmniAgent.
Workarounds
- If patching must wait, restrict network access to the affected service and place it behind authentication or a web application firewall rule.
CWE References
Technical References
This entry is built directly from NVD, CISA KEV and EPSS data. Verify fixed versions in the vendor advisory before you deploy.
Frequently asked questions
Related CVEs
In OpenClaw before 2026.2.23, tools.exec.safeBins validation for sort could be bypassed via GNU long-option abbreviations (such as --compress-prog) in allowlist mode, leading to approval-free execution paths that were intended to require approval. Only an exact string such as --compress-program was denied.
IBM API Connect 10.0.8.0 through 10.0.8.12 and 12.1.0.0 through 12.1.1.2 could allow a remote authenticated attacker to execute arbitrary code and access sensitive information from other tenants due to insufficient input validation. Uploaded custom module content is not sufficiently validated before it is processed, which could allow code to run with access beyond the user's own tenant.
Flowise is a drag & drop user interface to build a customized large language model flow. Prior to 3.1.3, Flowise validatePythonCodeForDataFrame in packages/components/src/pythonCodeValidator.ts can be bypassed with Unicode homoglyph identifiers, allowing arbitrary Python execution inside Pyodide and full OS command execution on the Flowise host via Pyodide js module interop. The validator gates pyodide.runPythonAsync in packages/components/nodes/agents/CSVAgent/CSVAgent.ts and packages/components/nodes/agents/AirtableAgent/AirtableAgent.ts with an ASCII word-boundary blacklist. JavaScript regex word boundaries are ASCII-only, while Python 3 NFKC-normalizes identifiers at parse time, so homoglyph forms such as __cl𝐚ss__, __subcl𝐚sses__, __b𝐚se__, and __b𝐮iltins__ bypass the blacklist and are parsed as their ASCII equivalents. This issue is fixed in version 3.1.3.
Craft is a content management system (CMS). Prior to 5.9.0-beta.1 and 4.17.0-beta.1, Craft CMS implements a blocklist to prevent potentially dangerous PHP functions from being called via Twig non-Closure arrow functions. In order to be able to successfully execute this attack, you need to either have allowAdminChanges enabled on production, or a compromised admin account, or an account with access to the System Messages utility. Several PHP functions are not included in the blocklist, which could allow malicious actors with the required permissions to execute various types of payloads, including RCEs, arbitrary file reads, SSRFs, and SSTIs. This vulnerability is fixed in 5.9.0-beta.1 and 4.17.0-beta.1.
Flowise is a drag & drop user interface to build a customized large language model flow. Prior to 3.1.0, the specific flaw exists within the run method of the CSV_Agents class. The issue results from the lack of proper sandboxing when evaluating an LLM generated python script. An attacker can leverage this vulnerability to execute code in the context of the user running the server. Using prompt injection techniques, an unauthenticated attacker with the ability to send prompts to a chatflow using the CSV Agent node may convince an LLM to respond with a malicious python script that executes attacker controlled commands on the Flowise server. This vulnerability is fixed in 3.1.0.
simple-git, an interface for running git commands in any node.js application, enables applications to execute Git operations from JavaScript. From 3.15.0 until 4.0.1, the default blockUnsafeOperationsPlugin does not classify trailer.<token>.cmd as unsafe configuration. An application that passes attacker-controlled values through SimpleGitOptions.config or inline -c arguments can therefore allow Git to invoke an attacker-selected shell command when git interpret-trailers processes the configured trailer. The command executes with the operating-system identity and permissions of the Node.js process. This issue is fixed in 4.0.1.
Earn your free CTEM certification.
Learn to scope, prioritize, validate and mobilize fixes for CVEs like CVE-2026-125368 with a free CTEM certification.
Beginner / Practitioner / Program Leader
Explore certifications