AI circle 'nuclear bomb level' security incident: LiteLLM, an open-source artifact with nearly 100 million monthly downloads, was' poisoned 'by hackers! What is LiteLLM? Why is it so important? Simply put, it is like the "universal translator and remote control" of AI models. There are now hundreds of large models such as OpenAI, Claude, Gemini, etc., each with a different interface. Programmers use LiteLLM to save time, as they only need to write one set of code to switch between different models at any time. Therefore, its monthly download volume is extremely high, and almost all mainstream AI proxy frameworks (such as DSPy, CrewAI) rely on it at the bottom layer. How do hackers' poison '? (Background and context) The hacker group TeamPCP played a "serial trick" this time: they first hacked into a commonly used code security scanning tool (Trivy). When the author of LiteLLM used this tool for "security checks", they actually stole the release token of PyPI (the official Python package management platform). On March 24th, hackers used tokens to directly release official versions 1.82.7 and 1.82.8 containing malicious code. Unreasonable destructive power: What can it steal? Once you install the toxic version, you don't need to do any extra action, and the malicious code (litellm_init-pth) will automatically execute covertly every time you run Python! It will crazily search for your AWS/Alibaba Cloud cloud credentials, all environment variables (including all your API Keys), SSH private keys, and digital currency wallets. This is known as a 'supply chain attack' in modern software engineering, and it is difficult to defend against! Dramatic discovery: 'Poor quality code' saved the entire industry. This super virus had been lurking for less than 4 hours before being exposed, and the reason was surprisingly: the hacker's code wrote bugs! Security researcher Callum McMahon automatically pulled the infected LiteLLM from the background while using the AI programmer cursor. As a result, the malicious code became trapped in a 'Fork Bomb', crazily generating child processes and directly bursting the computer's memory, causing it to crash. This only aroused vigilance and was reported and banned! Deep thinking on traffic: The "blind box crisis" of the AI era is now popular with "Vibe Coding" (using large models to generate and run code with just one click). We use various open-source dependency packages like building blocks, but we have no idea what's hidden inside. Traditional software engineering believes that "standing on the shoulders of giants" (dependency packages) is a good thing, but in today's AI frenzy, complex dependency trees are like blindly swallowing pills handed over by others. This is also why many experts are calling for reducing dependencies and returning to streamlined code. ️ Urgent self inspection and remedial suggestions: If you have installed or updated Python AI related frameworks before or after March 24th: Immediately check the version number: run pip show litellm. If it's 1.82.7 or 1.82.8, you've been hit! ⚠️ The entire development environment must be completely deleted and rebuilt! Urgently rotate and reset all passwords, API keys, SSH keys exposed on this machine!