LLM 0.37.2 fixes a security issue that could expose OpenAI API keys
Reported by letsdatascience and Simon Willison2 sources ↓
Simon Willison released LLM 0.37.2 to fix two cases where a configured OpenAI API key could be sent to an OpenAI-compatible endpoint from another provider.
Key points
- The issue affected use of LLM with OpenAI-compatible endpoints from other providers
- One case involved an API key name whose key had never been stored
- The other involved running a responses-enabled model with the chat_completions option
Sources
Details
Simon Willison’s release notes say the security issue was reported by Daniel de Araujo and is addressed in LLM 0.37.2. The release concerns LLM’s support for OpenAI-compatible API endpoints from other providers, configured through extra-openai-models.yaml or the `llm openai endpoint add` command.
The two reported cases were a key name configured through api_key_name for which no key had been stored, and running a model marked responses: true with the -o chat_completions 1 option. In either case, the configured OpenAI API key could be sent to the other provider.
Written from coverage as of