llm 0.37.2 fixes a leak that could send your OpenAI key to other providers
The point
Simon Willison released llm 0.37.2. The release fixes a security issue reported by Daniel de Araujo in GHSA-v9h7-jj4g-f862 and GHSA-r639-rr7r-x72w. The issue involves extra-openai-models.yaml, now managed by the llm openai endpoint add command in 0.37, for using LLM with OpenAI-compatible endpoints from other providers.
Why it matters
For builders, a stray key leak to a third party endpoint is a cost and security risk worth patching fast.
Key details
- Two leak paths: the key named by api_key_name had never been stored
- Or a responses: true model was run with -o chat_completions 1
Original ~2 min → 25 sec here Source: Simon Willison, Nov 11, 2026