On October 21, 2026, the open-source AI gateway LiteLLM was the target of a significant supply chain attack, its second this year. Attackers compromised the project's release pipeline and published a malicious version, v1.107.0, to the PyPI package repository and Docker Hub. The trojanized package was engineered to steal a wide range of sensitive credentials from the environments where it was deployed, including cloud provider logins, developer secrets, and API keys for services like OpenAI and Anthropic. The malicious versions were active for several hours before being removed. Users and organizations are advised that simply updating is insufficient; a full rotation of all potentially exposed credentials is required.
The attack represents a classic software supply chain compromise, where a trusted software distribution channel is used to deliver malware to downstream users. By compromising LiteLLM's release process, the attackers were able to push a malicious update that appeared legitimate. The malware's primary objective was credential harvesting, targeting high-value secrets commonly found in development and production environments that use AI services.
The malicious package was available during the following windows:
litellm==1.107.0): October 21, 2026, from 11:39 to 13:23 UTC.v1.107.0, latest, main-latest, main-stable): October 21, 2026, from 11:46 to 14:20 UTC.The last known safe version prior to the incident is v1.106.2.
The attack leveraged a Compromise Software Dependencies and Development Tools (T1195.001) technique. Once the malicious package was installed, it executed code designed to scan the environment for credentials. This included searching for:
AWS_ACCESS_KEY_ID, OPENAI_API_KEY).~/.aws/credentials).After harvesting the credentials, the malware would exfiltrate them to an attacker-controlled server. The package also included capabilities to load remote-code execution tools, allowing for potential second-stage attacks and persistence in the compromised environment.
The impact of this supply chain attack is potentially severe for any organization that downloaded and deployed the malicious version. LiteLLM is a popular tool used to manage and standardize interactions with various large language models (LLMs). As such, it is often deployed in environments with access to highly sensitive data and powerful API keys.
Compromised credentials could allow attackers to:
The incident underscores the significant risks associated with relying on open-source packages without proper security controls like dependency pinning and verification.
No specific C2 domains, IPs, or file hashes were provided in the source articles.
Security teams should hunt for evidence of the malicious LiteLLM version in their environments.
requirements.txt / pyproject.tomllitellm==1.107.0 to identify projects that may have installed the malicious version.CI/CD build logslitellm:1.107.0 or litellm:latest.CloudTrail, Azure Activity Logs1.107.0 or Docker images pulled with the affected tags during the incident window.v1.106.2 or older if an immediate update is not possible).latest tags or flexible version ranges, pin to a specific, known-good version hash (e.g., litellm==1.106.2). This prevents automatic updates to potentially malicious versions.New analysis reveals the LiteLLM supply chain attack included a backdoor for remote code execution, increasing the potential for broader corporate network compromise and arbitrary command execution.
Further analysis of the LiteLLM supply chain attack (version 1.107.0) confirms the presence of a sophisticated backdoor in addition to the infostealer. This backdoor allows threat actors to execute arbitrary commands on compromised developer machines, significantly increasing the risk beyond credential theft. The incident now maps to additional MITRE ATT&CK techniques including T1555.005 (Credentials from Password Stores), T1059.006 (Python execution), and T1048 (Exfiltration Over Alternative Medium). This expanded capability means attackers could pivot into corporate networks, inject malicious code into company software, and exfiltrate proprietary data, escalating the overall severity and potential for widespread organizational impact.
New details on LiteLLM attack: GitHub tokens used, malware persists via Python startup hooks, RAT backdoor found.
New analysis reveals attackers compromised LiteLLM's release pipeline using stolen GitHub CLI OAuth tokens from project engineers. The malicious package achieved persistence by modifying Python startup hooks (sitecustomize.py, usercustomize.py), ensuring execution whenever Python starts, not just on library import. Beyond credential harvesting, the malware also included a remote access trojan (RAT) backdoor. Microsoft and Datadog provided detailed technical analysis, confirming the sophisticated nature of the attack and updating detection guidance.
Malicious LiteLLM version v1.107.0 is published to PyPI.
Corresponding malicious Docker images are pushed to Docker Hub.
Malicious package is removed from PyPI.
Malicious Docker images are removed from Docker Hub.

Cybersecurity professional with over 10 years of specialized experience in security operations, threat intelligence, incident response, and security automation. Expertise spans SOAR/XSOAR orchestration, threat intelligence platforms, SIEM/UEBA analytics, and building cyber fusion centers. Background includes technical enablement, solution architecture for enterprise and government clients, and implementing security automation workflows across IR, TIP, and SOC use cases.
CyberNetSec.io uses automation to assist source monitoring, deduplication, observable extraction, and structured intelligence generation. Published analysis follows human-defined editorial standards and adds defensive context including MITRE ATT&CK, D3FEND, STIX, and Sigma where applicable. Read our editorial policy.
Help others stay informed about cybersecurity threats
Every tactic, technique, and sub-technique used in this threat has been identified and mapped to the MITRE ATT&CK framework for consistent, actionable threat language.
Observables and indicators of compromise (IOCs) have been extracted and cataloged. Risk has been assessed and correlated with known threat actors and historical campaigns.
Detection rules, incident response steps, and D3FEND-aligned mitigation strategies are included so your team can act on this intelligence immediately.
Structured threat data is packaged as a STIX 2.1 bundle and can be visualized as an interactive graph — relationships between actors, malware, techniques, and indicators.
Sigma detection rules are derived from the threat techniques in this article and can be converted for deployment across any major SIEM or EDR platform.