{"id":2665,"date":"2026-10-25T15:37:38","date_gmt":"2026-10-25T15:37:38","guid":{"rendered":"https:\/\/wyominginvest.com\/?p=2665"},"modified":"2026-10-25T15:37:38","modified_gmt":"2026-10-25T15:37:38","slug":"popular-llm-proxy-litellm-hit-by-supply-chain-attack-malicious-packages-compromise-thousands","status":"publish","type":"post","link":"https:\/\/wyominginvest.com\/?p=2665","title":{"rendered":"Popular LLM Proxy LiteLLM Hit by Supply Chain Attack, Malicious Packages Compromise Thousands"},"content":{"rendered":"<p>In a sophisticated supply chain attack that highlights the growing security risks within the AI development ecosystem, the popular open-source proxy tool LiteLLM was compromised, resulting in the distribution of malicious software to thousands of users. The breach, which occurred on October 21, saw attackers successfully inject malicious code into version 1.107.0 of the software, which is widely used to route requests across various Large Language Model (LLM) providers.<\/p>\n<p>The incident serves as a stark reminder of the fragility of modern software distribution pipelines. The malicious packages remained live on both the Python Package Index (PyPI) and Docker Hub for approximately three hours. During this narrow window, the compromised version was downloaded over 49,000 times, according to historical download data. The incident has since been remediated, with the malicious artifacts removed from PyPI and the affected Docker Hub tags either deleted or reverted to clean, verified images.<\/p>\n<h3>The Anatomy of the Breach<\/h3>\n<p>According to an official investigation published by the LiteLLM team, the intrusion began with the compromise of GitHub CLI OAuth tokens. These tokens belonged to an account within the project&#8217;s internal deployment\u2014an account that, while innocuously named, was inadvertently exposed to the public internet. <\/p>\n<p>The mechanism of the credential leak itself is particularly notable for its reliance on the burgeoning toolset used by AI developers. Investigators believe the tokens were inadvertently captured by &quot;Claude Code,&quot; an AI-driven coding assistant. Under specific operating conditions, the tool embedded these sensitive tokens into response bodies, which were subsequently captured in LiteLLM\u2019s internal logs. The attackers, having gained unauthorized access to the project&#8217;s infrastructure, were able to scrape these logs, retrieve the tokens, and use them to move laterally within the organization&#8217;s development environment.<\/p>\n<p>While LiteLLM has confirmed that the attackers successfully obtained three sets of developer GitHub CLI credentials, the exact method used to breach the internal deployment remains a subject of ongoing analysis. Once inside, the attackers moved quickly to subvert the software&#8217;s integrity. They utilized the stolen credentials to disable critical branch protection rules, allowing them to merge unauthorized changes into the codebase. <\/p>\n<p>These changes effectively redirected the project&#8217;s release process to a malicious source repository. By running a compromised GitHub workflow that pointed to a &quot;lookalike&quot; repository\u2014disguised as the legitimate <em>BerriAI\/litellm<\/em> project under the name <em>Berri-AI\/litellm<\/em>\u2014the attackers were able to sign and publish poisoned artifacts that appeared to be official releases.<\/p>\n<h3>Impact and Malicious Payload<\/h3>\n<p>The malicious version, 1.107.0, acted as a delivery mechanism for a multi-stage attack. Once installed, the package deployed a payload designed to function as both an information stealer and a Remote Access Trojan (RAT). Fortunately, LiteLLM has confirmed that its enterprise customers, who rely on private repository pulls rather than the public PyPI distribution, were shielded from this specific attack vector.<\/p>\n<p>The malware\u2019s primary objective was the wholesale exfiltration of sensitive data. Upon execution, the payload extracts three distinct files into the root of the user&#8217;s Python <code>site-packages<\/code> directory, ensuring that the malicious code is triggered automatically whenever a Python interpreter is initialized on the infected machine.<\/p>\n<p>The stealer component is highly efficient, designed to harvest a wide array of sensitive information from the developer\u2019s environment. This includes browser-stored credentials and cookies, developer-specific authentication tokens, environment variables, and cloud provider secrets for AWS, Google Cloud Platform (GCP), and Microsoft Azure. Furthermore, the malware targets API keys for prominent AI tools and services\u2014including Anthropic, GitHub Copilot, and OpenAI\u2014alongside SSH private keys, shell history files, and GPG keys. The stolen data is then encrypted and exfiltrated to a command-and-control server located at <em>litellm[.]dev<\/em>.<\/p>\n<h3>Post-Exploitation and LLMjacking<\/h3>\n<p>The reach of the attack extended beyond simple data theft. Cybersecurity firm Wiz, which analyzed the incident, reported that the attackers began leveraging the exfiltrated credentials within just 4.5 hours of the initial compromise. The researchers observed the attackers utilizing stolen AWS credentials to query Amazon Bedrock models, a technique commonly referred to as &quot;LLMjacking.&quot; This allows attackers to run expensive or unauthorized AI operations at the victim&#8217;s expense.<\/p>\n<p>In addition to the stealer, the RAT component provides the attackers with ongoing persistence. The malware maintains regular contact with a command-and-control server and utilizes a &quot;dead drop&quot; mechanism hosted on GitHub. By checking this dead drop every hour, the RAT can download and execute arbitrary Python code provided by the attackers, granting them full control over the compromised system and the ability to update their tactics in real-time.<\/p>\n<h3>Remediation and Industry Context<\/h3>\n<p>The security community has responded swiftly to the incident. Users who installed version 1.107.0 of LiteLLM are strongly urged to consider their systems compromised. The recommended remediation steps include the immediate removal of the affected version, a comprehensive rotation of all credentials that were present on the system during the infection window, and a thorough forensic sweep of the local environment for signs of residual malware. The most recent verified and clean version of the software is 1.106.2.<\/p>\n<p>This incident is part of a broader, concerning trend of supply chain attacks targeting the developer ecosystem. As the popularity of AI-driven development tools grows, so too does the interest of malicious actors in exploiting the pipelines that deliver these tools. Similar campaigns have recently targeted other widely used software components, such as the Rank Math PRO WordPress plugin and the AOMEI Backupper utility, highlighting that no segment of the software market is immune to these sophisticated threats.<\/p>\n<p>The LiteLLM breach underscores a critical shift in the threat landscape. Attackers are increasingly moving away from simple malware delivery toward sophisticated, automated credential harvesting that leverages the very tools intended to increase developer productivity. As the industry grapples with these challenges, the reliance on automated workflows and AI-assisted coding tools requires a parallel increase in security hygiene, particularly regarding the management of OAuth tokens, the securing of internal logs, and the strict enforcement of branch protections.<\/p>\n<p>The investigation into the LiteLLM compromise continues, with the project maintaining transparency regarding the incident to help the broader community harden its own infrastructure. For developers and organizations alike, the event serves as a sobering reminder that in the interconnected world of modern software development, a single misconfigured account or a stray log entry can provide the entry point for a large-scale supply chain compromise.<\/p>\n","protected":false},"excerpt":{"rendered":"<p>In a sophisticated supply chain attack that highlights the growing security risks within the AI development ecosystem, the popular open-source proxy tool LiteLLM was compromised, resulting in the distribution of malicious software to thousands of users. The breach, which occurred on October 21, saw attackers successfully inject malicious code into version 1.107.0 of the software,<\/p>\n","protected":false},"author":26,"featured_media":2664,"comment_status":"open","ping_status":"open","sticky":false,"template":"","format":"standard","meta":{"footnotes":""},"categories":[373],"tags":[246,374,368,245,144,2136,956,2137,1832,2135,5,670,1882,202],"class_list":["post-2665","post","type-post","status-publish","format-standard","has-post-thumbnail","hentry","category-bug-bounty-vulnerabilities","tag-attack","tag-bug-bounty","tag-chain","tag-compromise","tag-hacking","tag-litellm","tag-malicious","tag-packages","tag-popular","tag-proxy","tag-security","tag-supply","tag-thousands","tag-vulnerabilities"],"_links":{"self":[{"href":"https:\/\/wyominginvest.com\/index.php?rest_route=\/wp\/v2\/posts\/2665","targetHints":{"allow":["GET"]}}],"collection":[{"href":"https:\/\/wyominginvest.com\/index.php?rest_route=\/wp\/v2\/posts"}],"about":[{"href":"https:\/\/wyominginvest.com\/index.php?rest_route=\/wp\/v2\/types\/post"}],"author":[{"embeddable":true,"href":"https:\/\/wyominginvest.com\/index.php?rest_route=\/wp\/v2\/users\/26"}],"replies":[{"embeddable":true,"href":"https:\/\/wyominginvest.com\/index.php?rest_route=%2Fwp%2Fv2%2Fcomments&post=2665"}],"version-history":[{"count":0,"href":"https:\/\/wyominginvest.com\/index.php?rest_route=\/wp\/v2\/posts\/2665\/revisions"}],"wp:featuredmedia":[{"embeddable":true,"href":"https:\/\/wyominginvest.com\/index.php?rest_route=\/wp\/v2\/media\/2664"}],"wp:attachment":[{"href":"https:\/\/wyominginvest.com\/index.php?rest_route=%2Fwp%2Fv2%2Fmedia&parent=2665"}],"wp:term":[{"taxonomy":"category","embeddable":true,"href":"https:\/\/wyominginvest.com\/index.php?rest_route=%2Fwp%2Fv2%2Fcategories&post=2665"},{"taxonomy":"post_tag","embeddable":true,"href":"https:\/\/wyominginvest.com\/index.php?rest_route=%2Fwp%2Fv2%2Ftags&post=2665"}],"curies":[{"name":"wp","href":"https:\/\/api.w.org\/{rel}","templated":true}]}}