Your AI API keys might be the real product being harvested.
LiteLLM, a popular Python library for connecting applications to LLM providers, suffered a supply-chain attack after malicious version 1.107.0 was published on PyPI and Docker registries.
The malware could execute automatically when Python starts, even without explicitly importing LiteLLM.
What was at risk?
AI API keys from providers like OpenAI and Anthropic
Cloud credentials and environment secrets
SSH private keys and other sensitive authentication data
Remote code execution through a malicious loader
The attack also affected Docker tags including latest, main-latest, and main-stable while they pointed to compromised images.
What users should do:
• Avoid litellm==1.107.0 and affected images.
• Pin to 1.106.2 or another verified clean release.
• Rebuild affected environments from trusted artifacts.
• Rotate every credential the compromised machine could access.
LiteLLM’s maintainers confirmed the compromised artifacts were removed. If the malicious version ran in your environment, simply upgrading is not enough. Assume accessible secrets may have been exposed. <Cite refs={[“turn565655search0”,“turn565655search2”]}/>
The scary part? You don’t have to get hacked directly. Sometimes, you just install the wrong dependency.
How many AI projects are protecting their API keys, and how many are just hoping their dependencies are clean?
LiteLLM, a popular Python library for connecting applications to LLM providers, suffered a supply-chain attack after malicious version 1.107.0 was published on PyPI and Docker registries.
The malware could execute automatically when Python starts, even without explicitly importing LiteLLM.
What was at risk?
AI API keys from providers like OpenAI and Anthropic
Cloud credentials and environment secrets
SSH private keys and other sensitive authentication data
Remote code execution through a malicious loader
The attack also affected Docker tags including latest, main-latest, and main-stable while they pointed to compromised images.
What users should do:
• Avoid litellm==1.107.0 and affected images.
• Pin to 1.106.2 or another verified clean release.
• Rebuild affected environments from trusted artifacts.
• Rotate every credential the compromised machine could access.
LiteLLM’s maintainers confirmed the compromised artifacts were removed. If the malicious version ran in your environment, simply upgrading is not enough. Assume accessible secrets may have been exposed. <Cite refs={[“turn565655search0”,“turn565655search2”]}/>
The scary part? You don’t have to get hacked directly. Sometimes, you just install the wrong dependency.
How many AI projects are protecting their API keys, and how many are just hoping their dependencies are clean?