In an alarming development, Kaspersky’s Global Research and Analysis Team (GReAT) has uncovered a sophisticated supply chain attack that specifically targeted the Python Package Index (PyPI). This attack cleverly used malicious packages named ‘gptplus’ and ‘claudeai-eng’, which were designed to be mistaken for legitimate tools related to popular AI chatbot models like ChatGPT and Claude. By posing as useful utilities for these widely recognized AI platforms, the attackers aimed to deceive developers into downloading and using these compromised packages. Uploaded in November 2023, these malicious packages demonstrated seemingly legitimate functionalities, such as the ability to create chats and send messages to the language models, further aiding in their disguise.
As soon as these packages were imported, they activated a hidden process that executed Base64-encoded data. This process then downloaded a JavaUpdater.jar file from a GitHub repository. Within this file lay JarkaStealer, a powerful malware designed to harvest a wide array of data from compromised systems. Among the stolen information were browser data, screenshots, system information, and session data from applications like Telegram, Discord, and Steam. The deceptive packages managed to infiltrate over 1,700 systems across more than 30 countries. The highest download activity was observed in the United States, China, France, Germany, and Russia, highlighting the global reach and potential impact of this attack. Notably, the attackers did not appear to have any specific targets, making the attack even more concerning due to its indiscriminate nature.
Broader Implications and Continued Vigilance
Kaspersky’s Global Research and Analysis Team (GReAT) has discovered a sophisticated supply chain attack aimed at the Python Package Index (PyPI). Malicious packages named ‘gptplus’ and ‘claudeai-eng,’ designed to mimic legitimate tools linked to popular AI chatbots ChatGPT and Claude, were uploaded in November 2023. The attackers intended to deceive developers into downloading these compromised packages by making them appear as useful utilities.
Once imported, these packages began a hidden process that executed Base64-encoded data, subsequently downloading a JavaUpdater.jar file from a GitHub repository. This file contained JarkaStealer, potent malware designed to steal a wide range of data from affected systems. The data harvested included browser data, screenshots, system information, and session data from applications such as Telegram, Discord, and Steam.
These deceptive packages infiltrated over 1,700 systems in more than 30 countries, with the highest download activity in the United States, China, France, Germany, and Russia. The attack’s indiscriminate nature is particularly worrisome, as it targets no specific individuals or organizations, underscoring the potential global impact and reach of this threat.