Is Your AI Project Safe? Caution Against Malicious Python Packages

In a recent discovery by researchers from the Positive Technologies Expert Security Center (PT ESC), a malicious campaign targeted users of the Python Package Index (PyPI), raising concerns about the security of AI projects. The campaign involved two fraudulent packages, deepseeek and deepseekai, which aimed to exploit the increasing interest in AI and machine learning technologies. These packages were designed to extract sensitive user and system data under the guise of providing legitimate functionalities associated with DeepSeek AI clients. PyPI, widely used for accessing Python packages through package managers such as pip, pipenv, and poetry, unknowingly hosted these malicious entities, causing a potential threat to countless developers and users alike.

The fraudulent packages deepseeek and deepseekai were crafted to appear as genuine tools for interacting with DeepSeek AI services, luring users with promises of advanced text generation and completion features. However, their primary objective was much more sinister: collecting and transmitting sensitive information, including environment variables that often harbor crucial data such as API keys and database credentials. Once a user executed any commands associated with these packages, a malicious payload was activated, which in turn sent critical user and system information to a command-and-control (C2) server. Hosted on Pipedream, this C2 server facilitated the transmission of stolen details like user IDs, hostnames, and environment variables.

Discovery and Impact of the Attack

Upon detecting this malicious activity, PT ESC promptly notified PyPI administrators to take immediate action. Their swift response led to the removal of the malicious packages from the repository. However, despite their rapid intervention, the packages had already reached a significant number of users globally. Metrics revealed that deepseeek and deepseekai were downloaded 36 times via pip and the bandersnatch mirroring tool and 186 times through browsers and other methods across multiple countries. This widespread distribution underscored the global ramifications of the threat, highlighting the critical need for heightened vigilance and security practices among developers and users of open-source repositories.

The data extracted during the attack ranged from environment variables to user IDs, offering cybercriminals a treasure trove of sensitive information. Such data could be exploited for various illicit purposes, including unauthorized access to systems, hijacking of AI models, and exfiltration of proprietary data, thereby inflicting significant damage on affected individuals and organizations. The volume of downloads prior to detection also underlined the challenge of maintaining security within open-source ecosystems, where the fluidity and openness that drive innovation can sometimes become pathways for compromise.

Necessity for Caution and Future Measures

This incident serves as a stark reminder of the importance of vigilance and robust security measures when using open-source repositories. Developers and users must adopt best practices, such as thoroughly vetting packages, periodically reviewing dependencies, and staying informed about potential threats. Platforms like PyPI should also enhance their monitoring and response capabilities to swiftly detect and mitigate malicious activities. By fostering a collaborative effort among security researchers, platform administrators, and users, the community can better safeguard against such threats and maintain the integrity and trustworthiness of open-source ecosystems.

Explore more