Browse / AI Framework and Platform Flaws / Code and command execution
Code and command execution
133 records in AI Framework and Platform Flaws.
- NLTK dot2img runs Graphviz 'dot' without the trusted-exec chokepoint, enabling local code execution via a writable PATH element A partial earlier fix left one Graphviz call resolving dot from PATH without trust checks, so a planted binary in a writable directory runs.
- vLLM: Untrusted EPS inputs can trigger denial of service or Ghostscript RCE Image bytes from a request are autodetected by Pillow without a format allowlist, so an EPS file labeled as PNG reaches Ghostscript and runs commands.
- MLflow vulnerable to remote code execution on unauthenticated servers via third-party scorer deserialization importing caller-placed artifact modules On an unauthenticated MLflow server, scorer deserialisation imports a request-chosen module from an uploaded artifact and instantiates its class, running attacker code.
- PraisonAI: API deploy code generator embeds unescaped YAML fields into Python source Host and file path values from a YAML config are pasted unescaped into generated Python source, so injected expressions run when the API server starts.
- PraisonAI: Code Injection via f-string Interpolation in Deploy API Server Generation An unsanitised agents file value is interpolated into generated Python server code, which is then executed, so injected commands run.
- Langflow: Title Authenticated Remote Code Execution in validate_code via Malicious Decorators Description The code validator executes function definitions, but attacker-supplied decorators run at definition time, giving any logged-in user code execution.
- vm2: NodeVM `require.external` without an explicit `require.root` grants unrestricted host filesystem access and full RCE Default vm2 path settings let sandboxed code load the library through host require, then create an unrestricted instance that runs host OS commands.
- vm2: GHSA-m283-3h24-438v fix bypass leads to host RCE via call/apply indirection Calling Promise rejection handlers through call or apply indirection skips the sanitiser, letting a host error carrying a host object reach the sandbox and run commands.
- Cline: Cross-Origin WebSocket Hijacking in Cline Hub Dashboard (`/browser` endpoint) A malicious website opens a cross-origin WebSocket to the local dashboard, which accepts it without an Origin check and issues commands that auto-approved tools execute.
- n8n: Expression Sandbox Escape via Class-Field Sanitizer Rebinding Can Lead to Code Execution A class field named __sanitize rebinds the expression sanitizer's this, reaching the Function constructor and running arbitrary JavaScript in n8n.
- NLTK: Uncontrolled search path when invoking the Graphviz 'dot' binary The dot binary is launched by bare name, so a planted file named dot in the working directory or a PATH entry runs instead of Graphviz.
- NLTK: JVM argument injection bypass via per-call options in the NLTK Stanford wrappers (incomplete fix of CVE-2026-12841) Per-call Java options skip the validator added for an earlier fix, letting callers pass agent flags that run code inside the JVM.
- LiteLLM vulnerable to server-side template injection in the /prompts/test endpoint The prompt test endpoint renders attacker-supplied template text with an unsandboxed Jinja2 environment, letting unauthenticated callers run OS commands.
- Flowise RCE via TypeORM DataSource Authenticated users can pass arbitrary options into a TypeORM DataSource through node settings, which can be abused to execute code on the server.
- Flowise RCE via SQLite Record Manager Node The SQLite record manager node lets additionalConfig override its database path, enabling attacker-chosen file writes that lead to code execution.
- NLTK vulnerable to Eval Injection via collocations CLI arguments Command-line arguments are passed straight into eval as attribute names, letting an attacker inject Python expressions that run OS commands.
- Open WebUI: Cross-user code-interpreter and tool execution via unvalidated Socket.IO event-caller session_id A user supplies another user's socket session id, so code and tool calls run in the victim's session and can reach admin-only server-side function execution.
- LiteLLM: Custom Code Guardrails production endpoints bypass code safety checks Production create and update routes for custom guardrail code skip the sandbox used by the test endpoint, so submitted Python runs inside the proxy.
- n8n: Race Condition in Git Clone Node Allows Authenticated Users to Achieve Remote Code Execution A race between validating the clone path and running it lets a user swap in a symlink, planting a repository where the server loads custom nodes as code.
- n8n: Legacy Expression Evaluator Sanitizer Bypass Leads to Authenticated Code Execution An authenticated workflow editor can get past the legacy expression evaluator's member-access sanitizer and run code as the server process.
- n8n: Authenticated code execution in the n8n Git node A workflow user supplies a crafted local Git repository whose hooks run when the Git node operates on it, executing commands as the server process.
- Langflow: Code Injection Vulnerability in Code Validation Endpoint The code validation endpoint lets authenticated users inject Python that the server executes, allowing OS commands and reading of sensitive files.
- Langflow: Unauthenticated RCE in Shareable Playgrounds A public flow endpoint accepts replacement Python code in node definitions and runs it without authentication.
- Open WebUI: Jupyter code execution works despite `ENABLE_CODE_EXECUTION=false` — feature gate bypassed The Open WebUI code execution endpoint forwards Python to Jupyter without checking the ENABLE_CODE_EXECUTION setting, so any verified user can run code.
- FlowiseAI: Authenticated Host RCE via POST /api/v1/node-custom-function and NodeVM Sandbox Escape A signed-in user posts JavaScript to a custom function endpoint with no route permission check, then escapes the NodeVM sandbox to reach host child_process.
- n8n: HTTP Request Node Pagination Prototype Pollution to RCE A user-controlled pagination parameter in the HTTP Request node pollutes global object prototypes, which can be chained into code execution.
- n8n Has an XML Node Prototype Pollution Patch Bypass A prior fix for prototype pollution in the XML node can be circumvented by a workflow author, and combined with other nodes it reaches code execution on the host.
- Open WebUI: Missing `workspace.tools` Authorization Check on Tool Update Endpoint Allows Privilege Escalation to Code Execution The tool update endpoint skips the workspace permission check that tool creation enforces, so a restricted user can replace a tool's Python code and run it.
- BentoML Dockerfile command injection via docker.base_image (sister of pending GHSA-w2pm-x38x-jp44 / CVE-2026-33744 / CVE-2026-35043) A multi-line base image value in the build config injects extra Dockerfile instructions, which then run as RUN commands during the docker build step.
- Open WebUI has a CORS misconfiguration and session validation issue Permissive cross-origin rules and weak session validation let a malicious page take a logged-in admin's session and reach remote code execution with one click.
- n8n has XML Node Prototype Pollution that to RCE XML node parsing lets an authenticated workflow author pollute global object prototypes, which can chain with other nodes into remote code execution.
- n8n has Prototype Pollution in XML Webhook Body Parser that Leads to RCE Crafted XML webhook bodies pollute JavaScript object prototypes in the parser, and the pollution is chained with the Git node to run commands on the host.
- LiteLLM: Server-Side Template Injection in /prompts/test endpoint A user-supplied prompt template is rendered without a sandbox, so crafted template expressions can execute code inside the proxy process.
- H2O-3 is Vulnerable to Code Injection A parameter blocklist meant for MySQL JDBC is avoided by switching to the PostgreSQL driver, whose socket factory options run attacker-supplied code.
- Flowise: Code Injection in CSVAgent leads to Authenticated RCE Unsanitised custom Pandas read code is interpolated into Python run by pyodide in the CSV agent, so a crafted chatflow executes OS commands.
- Flowise: Parameter Override Bypass Remote Command Execution A substring check meant for file inputs accepts the keyword anywhere in a request, letting unauthenticated users override protected config such as NODE_OPTIONS and run commands.
- Flowise: File Upload Validation Bypass in createAttachment Adding a JavaScript MIME type to chatflow upload settings lets an attacker store a Node web shell that then runs on the server.
- Google Agent Development Kit (ADK) has a Code Injection and Missing Authentication vulnerability An unauthenticated caller can reach a route in the ADK server that runs attacker-supplied code on the host.
- MetaGPT has an eval injection via a cross-site request forgery attack A cross-site request forgery against the Mineflayer HTTP API triggers evaluation of attacker-supplied code in the agent's environment.
- FoundationAgents MetaGPT vulnerable to OS Command Injection in metagpt/utils/common.py The MIME type helper builds an operating system command from untrusted input, allowing injected shell commands to execute.
- PraisonAI Vulnerable to Remote Code Execution via YAML Deserialization in Agent Definition Loading A crafted agent YAML file uploaded through the API is parsed with an unsafe loader that evaluates embedded JavaScript functions.
- BentoML: Command Injection in cloud deployment setup script Unquoted system_packages values from bentofile.yaml are placed in a setup script run on cloud build infrastructure, a path an earlier quoting fix missed.
- Kedro has Arbitrary Code Execution via Malicious Logging Configuration Setting the logging config path to a file containing a special factory key makes dictConfig call an arbitrary callable at application startup.
- Mlflow: Command Injection when serving models with enable_mlserver=True The model URI is interpolated unescaped into a bash -c command, so shell command substitutions inside it run when the model is served.
- BentoML has Dockerfile Command Injection via system_packages in bentofile.yaml The system_packages values in bentofile.yaml are placed unquoted into Dockerfile RUN commands, so shell metacharacters in a package name run commands during the build.
- n8n: Prototype Pollution in XML and GSuiteAdmin node parameters lead to RCE A crafted node parameter in a workflow pollutes Object.prototype, and the pollution can be escalated to code execution on the server.
- Unauthenticated Remote Code Execution in Langflow via Public Flow Build Endpoint An unauthenticated public build endpoint accepts attacker-supplied flow data whose node definitions contain Python that the server runs with exec.
- MLflow has a command injection in mlflow/sagemaker/__init__.py The sagemaker CLI interpolates the container image name into a shell command run through os.system, so shell metacharacters in the flag execute.
- SGLang's multimodal generation module is vulnerable to unauthenticated remote code execution through the ZMQ broker The ZMQ broker in SGLang's multimodal module unpickles network messages without authentication, so a remote sender can run code.
- SageMaker Python SDK replaced eval() with safe parser in JumpStart search functionality The search_hub query parameter is passed to eval(), so crafted input in the JumpStart search path can run arbitrary Python.