litellm ai-safety developer-tools credential-theft

LiteLLM Python package compromised by supply-chain attack

[Security]: CRITICAL: Malicious litellm_init.pth in litellm 1.82.8 — credential stealer · Issue #24512 · BerriAI/litellm

github.com

March 24, 2026

3 min read

🔥🔥🔥🔥🔥

72/100

Summary

The litellm 1.82.8 package on PyPI contains a malicious litellm_init.pth file that executes a credential-stealing script upon starting the Python interpreter. Users are advised to avoid this version to prevent credential theft.

Key Takeaways

The litellm==1.82.8 PyPI package contains a malicious .pth file that executes a credential-stealing script upon Python interpreter startup without requiring an import statement.
The malicious script collects sensitive information from the host system, including API keys, SSH keys, Git credentials, and various cloud provider credentials.
Collected data is encrypted and exfiltrated to the domain models.litellm.cloud using a POST request.
The attack exploits the automatic execution of .pth files in Python's site-packages directory, making it stealthy and difficult to detect.

Read original article

View story →

Coverage & discussions

Hacker News · github.com

LiteLLM Python package compromised by supply-chain attack ↗

Community discussions

Hacker News discussion ↗

Historical count: 5 · 5 saved in sample · Historical sample

Analysis ready

Community Sentiment

Mixed

Positives

The LiteLLM team is actively investigating the supply-chain attack, indicating a commitment to transparency and improvement in their security practices.
The quarantine of the compromised package on PyPI demonstrates proactive measures to protect users from potential threats.

Concerns

The reliance on dependencies and development setups raises significant trust issues, highlighting the need for better isolation and security in AI development environments.
Concerns about training data integrity suggest that compromised code could inadvertently influence AI models, leading to potential security vulnerabilities in their outputs.