Skip to content
Catalyst Note CN markCNCatalystNote

Noting the changes that set others in motion.

Model updates

LLM 0.37.2 fixes a security issue in OpenAI-compatible endpoint handling

Version 0.37.2 fixes a security issue that could accidentally send your configured OpenAI API key to other providers when using LLM with OpenAI-compatible API endpoints.

LLM 0.37.2 fixes a security issue in OpenAI-compatible endpoint handling
Depth
1 outlet · newest source 114h ago

Key points

  • Release 0.37.2 addresses security issues GHSA-v9h7-jj4g-f862 and GHSA-r639-rr7r-x72w.
  • Daniel de Araujo reported the vulnerabilities in the extra-openai-models.yaml mechanism.
  • The leak occurred if a named key was missing or when running responses models with chat_completions.

The release addresses flaws identified by Daniel de Araujo in GHSA-v9h7-jj4g-f862 and GHSA-r639-rr7r-x72w. These vulnerabilities involved the extra-openai-models.yaml mechanism, which allows LLM to use OpenAI-compatible API endpoints from other providers. The feature is now managed by the llm openai endpoint add command, introduced in version 0.37.

Evidence: reportedllm 0.37.2

aifleet-willison

Evidence: reportedllm 0.37.2

aifleet-willison

The claim

Daniel de Araujo reported two scenarios where the tool could accidentally send a configured OpenAI API key to a third-party provider. The first scenario occurred if the key named by api_key_name had never been stored. The second scenario involved running a model with responses: true using the -o chat_completions flag. This behavior exposed credentials to unintended destinations during API calls.

Evidence: reportedllm 0.37.2

aifleet-willison

The response

This change aims to prevent accidental credential leakage when using OpenAI-compatible endpoints from other providers. No further details on the specific code changes were provided in the report.

Why this matters now

Users of LLM who configure third-party OpenAI-compatible endpoints are affected. The vulnerability existed in the mechanism for handling extra models. Version 0.37.2 is the first release to address these specific credential-leak scenarios.

Evidence: reportedllm 0.37.2

aifleet-willison

What buyers should weigh

Enterprise IT buyers should assess their current LLM version. If using versions prior to 0.37.2, there is a risk of API key exposure to third-party providers. Buyers must decide whether to upgrade immediately or audit their endpoint configurations for potential leaks.

Evidence: reportedllm 0.37.2

aifleet-willison

Why it matters

  • The update addresses a security flaw that could leak OpenAI API keys to third-party providers.
  • The issue arises when an api_key_name has not been previously stored in the configuration.
  • Running responses: true models with the -o chat_completions flag also triggered the leak.
  • The fix resolves vulnerabilities identified in GHSA-v9h7-jj4g-f862 and GHSA-r639-rr7r-x72w.

Topics: llm, security, API keys, releases, OpenAI

Updated

Ask the article

Local keyword search of this article; no generated answers.

Did this article give you the evidence you needed?

No vote saved on this device.
Your choice stays on this device. No public tally.

Terms in this story

LLM

A command-line tool for interacting with large language models.

extra-openai-models.yaml

A configuration file for adding OpenAI-compatible API endpoints from other providers.

GHSA

GitHub Security Advisory, a standard for reporting software vulnerabilities.

Questions readers ask

How did the API key leak occur?

It occurred if a named key was missing or when running responses models with the chat_completions flag.

according to the published report, these vulnerabilities?

Daniel de Araujo reported the issues in GHSA-v9h7-jj4g-f862 and GHSA-r639-rr7r-x72w.

Connected stories

This storyOpenAI Chief Ranks UAE First in AI Adoption Amid New Deployment DealOpenAI Chief Ranks UAE First in ChatGPT desktop app combines Chat and Work tabs with unified token pool for testersChatGPT desktop app combines ChaRust NYC Report Highlights Reflections on Rust in the LLM AgeRust NYC Report Highlights RefleNvidia CEO admits relationship with woman U.S. intelligence says was recruited by China’s top spy agencyNvidia CEO admits relationship w

Reading receipt

  • Release 0.37.2 addresses security issues GHSA-v9h7-jj4g-f862 and GHSA-r639-rr7r-x72w.
  • Daniel de Araujo reported the vulnerabilities in the extra-openai-models.yaml mechanism.
  • The leak occurred if a named key was missing or when running responses models with chat_completions.
Time spent: 0:00 · Depth: Full

Sources

  1. Simon Willison AI

Related stories

Background reading