LLM CLI tool bug could leak OpenAI API key to third-party endpoints

A vulnerability in Simon Willison's 'llm' command-line tool, reported by Daniel de Araujo, could cause a user's configured OpenAI API key to be sent to other OpenAI-compatible providers under certain misconfiguration conditions involving extra-openai-models.yaml. The issue was fixed in llm 0.37.2.

Disclosed 11 November 2026 · Record updated 12 November 2026

Impact

OpenAI API keys could be inadvertently sent to third-party OpenAI-compatible API providers if certain configuration conditions were met.

Our coverage

A lockbox labeled with a key symbol springs open, spilling golden tokens that scatter across a desk toward shadowy third-party hands reaching from the darkness.

Vulnerabilities

Fix for llm tool stops OpenAI keys leaking to other providers

Version 0.37.2 of the open-source command-line tool patches two cases where a user's configured OpenAI key could be sent to a third-party endpoint.

12 Nov 2026

Sources

simonwillison.nethttps://simonwillison.net/2026/Nov/11/llm