llm 0.37.2 patches two paths that could send OpenAI API keys to other providers

Simon Willison has released llm 0.37.2, a patch version of his command-line tool for accessing large language models. The release fixes two security issues, tracked as GHSA-v9h7-jj4g-f862 and GHSA-r639-rr7r-x72w, that were reported by Daniel de Araujo.

LLM connects to OpenAI-compatible API endpoints from other providers through a file called extra-openai-models.yaml. Version 0.37 added the llm openai endpoint add command, which manages that file.

Willison says de Araujo found two ways a configured OpenAI API key could be sent by accident to one of those providers. One happened when the key named by api_key_name had never been stored. The other happened when a model with responses: true was run with -o chat_completions 1.

Developers who use LLM with third-party OpenAI-compatible endpoints should upgrade to 0.37.2.

Agents that rely on api_key_name or on chat_completions routing should check their configuration, since the affected paths could have sent an OpenAI key to the wrong provider.