LiteLLM proxy from 1.81.8: a sandbox escape attackers are exploiting runs code, update to 1.83.11
Anyone running the LiteLLM proxy server to route requests to AI models, especially one others can reach.
How to fix it
You haveUpdate to
>= 1.81.8, < 1.83.10→1.83.11
Check your version
Which version you havepip show litellm
Update
pippip install --upgrade "litellm>=1.83.11"
uv add "litellm>=1.83.11"
If you can't update yet
LiteLLM says to block POST /guardrails/test_custom_code at your reverse proxy or API gateway until you can upgrade.
Someone with a proxy-admin key can break out of the custom-code guardrail sandbox and run code in the proxy, which runs as root in the default Docker image. LiteLLM's version data lists 1.83.10 as patched, but its advisory text says the fix is in 1.83.11, so take 1.83.11 or later.