Vulnerabilities > CVE-2023-37274 - Code Injection vulnerability in Agpt Auto-Gpt
Summary
Auto-GPT is an experimental open-source application showcasing the capabilities of the GPT-4 language model. When Auto-GPT is executed directly on the host system via the provided run.sh or run.bat files, custom Python code execution is sandboxed using a temporary dedicated docker container which should not have access to any files outside of the Auto-GPT workspace directory. Before v0.4.3, the `execute_python_code` command (introduced in v0.4.1) does not sanitize the `basename` arg before writing LLM-supplied code to a file with an LLM-supplied name. This allows for a path traversal attack that can overwrite any .py file outside the workspace directory by specifying a `basename` such as `../../../main.py`. This can further be abused to achieve arbitrary code execution on the host running Auto-GPT by e.g. overwriting autogpt/main.py which will be executed outside of the docker environment meant to sandbox custom python code execution the next time Auto-GPT is started. The issue has been patched in version 0.4.3. As a workaround, the risk introduced by this vulnerability can be remediated by running Auto-GPT in a virtual machine, or another environment in which damage to files or corruption of the program is not a critical problem.
Vulnerable Configurations
| Part | Description | Count |
|---|---|---|
| Application | 1 |
Common Weakness Enumeration (CWE)
Common Attack Pattern Enumeration and Classification (CAPEC)
- Leverage Executable Code in Non-Executable Files An attack of this type exploits a system's trust in configuration and resource files, when the executable loads the resource (such as an image file or configuration file) the attacker has modified the file to either execute malicious code directly or manipulate the target process (e.g. application server) to execute based on the malicious configuration parameters. Since systems are increasingly interrelated mashing up resources from local and remote sources the possibility of this attack occurring is high. The attack can be directed at a client system, such as causing buffer overrun through loading seemingly benign image files, as in Microsoft Security Bulletin MS04-028 where specially crafted JPEG files could cause a buffer overrun once loaded into the browser. Another example targets clients reading pdf files. In this case the attacker simply appends javascript to the end of a legitimate url for a pdf (http://www.gnucitizen.org/blog/danger-danger-danger/) http://path/to/pdf/file.pdf#whatever_name_you_want=javascript:your_code_here The client assumes that they are reading a pdf, but the attacker has modified the resource and loaded executable javascript into the client's browser process. The attack can also target server processes. The attacker edits the resource or configuration file, for example a web.xml file used to configure security permissions for a J2EE app server, adding role name "public" grants all users with the public role the ability to use the administration functionality. The server trusts its configuration file to be correct, but when they are manipulated, the attacker gains full control.
- Manipulating User-Controlled Variables This attack targets user controlled variables (DEBUG=1, PHP Globals, and So Forth). An attacker can override environment variables leveraging user-supplied, untrusted query variables directly used on the application server without any data sanitization. In extreme cases, the attacker can change variables controlling the business logic of the application. For instance, in languages like PHP, a number of poorly set default configurations may allow the user to override variables.