Attack ChainAIDerived
Hypothetical attack flow derived from CVE metadata – not a confirmed exploit chain
-
AccessRegister or authenticate as low-privileged user
-
DeliveryCreate workflow with python code node
-
ExploitBypass regex import filter via __import__('subprocess')
-
ExecutionExecute OS commands in unsandboxed subprocess
-
ImpactExfiltrate env vars, API keys, and pivot to MySQL/MinIO/Redis/etcd
Vulnerability Assessment
| Weakness |
CWE
ASVS 5.0
V13 - Configuration Secure default settings
|
| Exploitation Conditions | Requires (1) an authenticated account - any registered user suffices per PR:L, with no admin role needed - and (2) the vulnerable default configuration CODE_RUNNER_TYPE="local" (shipped default in docker/.env.example before 0.5.0), which runs workflow code nodes in an unsandboxed python3 subprocess on the host. The attacker must be able to create/run a workflow code node, and the regex import filter is defeated using __import__('subprocess'). Exploitation is LIMITED by the authentication requirement and by deployments that have changed the runner to "sandbox" (the patched default) or otherwise isolated the code runner; instances running the fixed default or without the local runner enabled are not exploitable via this path. No user interaction is required (UI:N). |
| Impact | This is a credible, high-priority risk for any organization running a self-hosted Coze Studio instance prior to 0.5.0. The vendor-supplied CVSS 4.0 vector (AV:N/AC:L/AT:N/PR:L/UI:N) with VC:H/VI:H/VA:H scores 8.7 and is consistent with the description: a network-reachable, low-complexity attack requiring only low privileges (PR:L = any registered user) and no user interaction, yielding full confidentiality, integrity and availability impact via arbitrary command execution. Publicly available exploit code exists (Lasso Security), which materially raises near-term risk, but this must NOT be conflated with active exploitation - the CVE is not listed in CISA KEV and no EPSS probability was provided in the input, so widespread opportunistic exploitation cannot be confirmed from available data. The root cause is an insecure default (CWE-1188): the shipped CODE_RUNNER_TYPE="local" setting runs workflow code nodes in an unsandboxed python3 subprocess, and the regex-based import filter is trivially bypassed via __import__('subprocess'). The main mitigating factor is that exploitation requires an authenticated account on the instance and the default (unpatched) local runner configuration; it is not a pre-auth internet-wide worm candidate. Net assessment: a genuine priority for exposed/multi-tenant deployments, somewhat tempered by the authentication requirement and the product's limited production footprint rather than being a high-CVSS-but-low-real-risk issue. |
| Exploit ScenarioDerived | An attacker who registers (or already holds) a low-privileged account on a default-configured Coze Studio instance builds a workflow containing a python code node and bypasses the import filter with __import__('subprocess') to run OS commands in the unsandboxed runner. From there they read environment variables and model API keys and pivot to the backing MySQL, MinIO, Redis and etcd services. A public proof-of-concept exists, and the network/low-complexity vector makes exploitation straightforward once an account is obtained. |
Get an e-mail when this CVE is exploited, patched or re-scored. Free account.
Alert meAnalysisAIv1 – Oct 20, 2026
Remote code execution in Coze Studio before 0.5.0 allows any registered user to run arbitrary commands through workflow code nodes because the shipped default CODE_RUNNER_TYPE="local" executes Python in an unsandboxed host subprocess. The regex import filter can be bypassed with __import__('subprocess'), exposing environment variables, model API keys, and connected services including MySQL, MinIO, Redis, and etcd; publicly available exploit code exists. The vendor fix changes the default runner to "sandbox" in 0.5.0, so self-hosted instances that have not overridden the default are the primary exposure.
RemediationAI
Upgrade Coze Studio to 0.5.0 or later, which changes the default CODE_RUNNER_TYPE from "local" to "sandbox" per commit fcf40d1008c100108c23be051a49950283f2c2cf. If immediate upgrade is not possible, set CODE_RUNNER_TYPE="sandbox" in docker/.env and redeploy or restart the affected services, then verify workflow code nodes no longer spawn a host python3 process; this may change or limit code-node behavior and should be tested. As compensating controls, restrict registration and workflow-authoring access to trusted users (trade-off: reduced self-service), rotate model API keys and any credentials reachable from the Coze Studio host, and restrict the host's network access to MySQL, MinIO, Redis, and etcd so command execution cannot pivot to those services. Monitor for suspicious subprocess activity and review the vendor advisory at https://www.vulncheck.com/advisories/coze-studio-before-0.5.0-rce-via-unsandboxed-local-python-code-runner and the patch commit at https://github.com/coze-dev/coze-studio/commit/fcf40d1008c100108c23be051a49950283f2c2cf.
Affected VersionsENISA
Affected Packages
Source: OSV (Open Source Vulnerabilities)
Affects your stack
Affects 1 package in euvd
Checking this CVE against your own stacks and servers is part of the Basic plan and up.
Technical ContextAI · show details
Coze Studio is a self-hosted AI agent and workflow platform. Its workflow code nodes are intended to run user-supplied Python under a configurable code runner. The vulnerable configuration, shipped in docker/.env.example before 0.5.0 at the referenced line 234, sets CODE_RUNNER_TYPE="local", which launches python3 as a host subprocess rather than in a sandboxed execution environment. The root cause class is CWE-1188 (Insecure Default Initialization of Resource): a security-sensitive default is unsafe out of the box. The application attempts to block dangerous imports with a regex filter, but __import__('subprocess') evades that filter and grants command execution. The affected CPE is cpe:2.3:a:coze-dev:coze-studio:*:*:*:*:*:*:*:*. Exploitation requires an authenticated account (CVSS 4.0 PR:L; assessed 3.1 AV:N/AC:L/PR:L/UI:N/S:U/C:H/I:H/A:H) and the local runner default; deployments already set to sandbox or upgraded to 0.5.0 are not exploitable via this path. The patch commit fcf40d1008c100108c23be051a49950283f2c2cf changes the example default to "sandbox".
Know when Coze Studio gets its next vulnerability
- Watch products and get an email when something new is published
- Check this CVE against your own stack and servers
- Exploit and PoC references, threat intelligence from 22 sources
No credit card · 7-day full trial
Severity by source
CVSS:4.0/AV:N/AC:L/AT:N/PR:L/UI:N/VC:H/VI:H/VA:H/SC:N/SI:N/SA:N
Network-reachable code node needs only a registered account so PR:L and AC:L; unsandboxed command execution gives full C/I/A; scope kept S:U to align with the vendor's SC:N/SI:N/SA:N.
Primary rating from Vendor (VulnCheck).
CVSS VectorVendor: VulnCheck
DescriptionCVE.org
Coze Studio before 0.5.0 contains an insecure default configuration that allows any registered user to execute commands by running workflow code nodes in an unsandboxed python3 subprocess. Attackers can bypass the regex import filter with __import__('subprocess') to run commands, stealing environment variables, model API keys, and accessing MySQL, MinIO, Redis and etcd.
Articles & Coverage 1
Technical POC Analysis
CWE-1188: Insecure Default Initialization of Resource.
Full technical POC analysis - root cause, step-by-step exploitation and patch breakdown - is part of the Pro and Business plans.
Executive ViewAI
Coze Studio, the self-hosted AI agent/workflow platform, ships with a default code-runner setting that lets any user with a registered account execute arbitrary commands on the host, exposing environment variables, model API keys, and connected MySQL, MinIO, Redis, and etcd services. Because exploit code is publicly available and many self-hosted deployments never change the shipped default, leadership should treat every unmodified instance as an active credential-compromise risk rather than a theoretical code-quality issue. Impact concentrates on data and service confidentiality across the AI platform and its backing infrastructure, not just on the workflow feature itself.
Apply vendor patch. Until Coze Studio 0.5.0 is deployed, set CODE_RUNNER_TYPE="sandbox" (or an isolated remote runner) instead of the shipped "local" default, isolate affected hosts from production databases and object storage, restrict account registration and workflow-code authoring to trusted users, and rotate all secrets reachable from the runner.
Threat Intelligence Overview
How this decision was reached
The engine reads each signal below and applies a fixed list of rules from top to bottom. The first rule that matches sets the recommended action – like a statement, every line is shown.
What every action level means
SSVC AssessmentENISA
Priority Score
Related Vulnerabilities in Coze Studio RCE and IDOR Vulnerabilities
View all 2More in Coze Studio
View allImproper authorization in Coze Studio through 0.5.1 lets an authenticated user (CVSS PR:L) abuse the workflow SQL custom
Server-side request forgery in Coze Studio through version 0.5.1 allows an authenticated user who can register or config
SQL injection in ByteDance coze-studio up to version 0.5.1 allows authenticated remote attackers to manipulate the Execu
EUVD-2026-104690
GHSA-46jr-m373-8hpx
NVD
CVE.org
EPSS
Vulnrichment
Exploit-DB