AI Sec Watch

CVE-2024-4099 is a vulnerability in GitLab EE (a Git repository management tool) affecting versions 16.0-17.2.7, 17.3-17.3.3, and 17.4-17.4.0 where an AI feature failed to clean up unsanitized input, potentially allowing attackers to perform prompt injection (tricking the AI by hiding instructions in its input). The vulnerability has a CVSS score (a 0-10 severity rating) of 4.0, indicating low to moderate severity.

Monica AI Assistant desktop application v2.3.0 has a vulnerability where attackers can use prompt injection (tricking an AI by hiding instructions in its input) with a specially crafted image to steal sensitive chat data from the current session and send it to an attacker-controlled server. This flaw allows unauthorized people to access private information from users' conversations.

The Chatbot with ChatGPT WordPress plugin before version 2.4.6 has a missing authorization flaw in one of its REST endpoints (a web interface for accessing the plugin's functions), which allows unauthenticated users (anyone without login credentials) to retrieve and decode an OpenAI API key (a secret credential that grants access to OpenAI's services). This vulnerability exposes the API key to attackers.

CVE-2024-40442 is a privilege escalation vulnerability (a security flaw where an attacker gains higher access levels than they should have) in Doccano v.1.8.4 and its Auto Labeling Pipeline module v.0.1.23. A remote attacker can exploit this weakness by sending a specially crafted REST request (a malicious command sent over the web), which involves improper code injection (inserting malicious code into the system).

CVE-2024-40441 is a privilege escalation vulnerability (a bug that lets attackers gain higher-level access than they should have) in Doccano v.1.8.4, an open source tool for labeling data to train machine learning models, and its Auto Labeling Pipeline module v.0.1.23. A remote attacker can exploit this by manipulating the model_attribs parameter to escalate their privileges.

Attackers can inject spyware into ChatGPT's memory (a feature that stores information across chat sessions) through prompt injection (tricking an AI by hiding instructions in its input) on untrusted websites, allowing them to continuously steal everything a user types in future conversations. The vulnerability exploits a weakness where a security check called url_safe was performed only on the user's device rather than on OpenAI's servers, and becomes more dangerous when combined with the Memory feature that persists attacker-controlled instructions. OpenAI released a fix for the macOS app, and users should update to the latest version.

LangChain Experimental versions 0.1.17 through 0.3.0 contain a vulnerability that allows attackers to execute arbitrary code (run malicious commands on a system) through a component called LLMSymbolicMathChain, which uses sympy.sympify (a function that evaluates mathematical expressions in an unsafe way). The root cause is improper input validation (failing to check that user input is safe before processing it).

A vulnerability in the ilab model serve component allows attackers to cause a Denial of Service (DoS, where a service becomes unavailable to legitimate users) by sending a large value for the best_of parameter to the vllm JSON web API (a web interface for accessing an LLM). The API doesn't properly manage timeouts or resource limits, so an attacker can exhaust system resources and crash the service.

CVE-2024-8768 is a bug in vLLM (a library for running large language models) where sending an API request with an empty prompt crashes the server, causing a denial of service (making the service unavailable to users). The flaw is classified as a reachable assertion vulnerability, meaning the code hits an unexpected condition it wasn't designed to handle.