cyber.hendryadrian.com
Security Report · wiz.io

Wiz Finds Verified GitHub Secret Leaks at 65% of AI 50 Companies

· Original article ↗

Summary

Wiz says its GitHub scan found verified secret leaks at 65% of analyzed Forbes AI 50 companies, including credentials in deleted forks, gists, workflow logs, and developers’ public repositories.

Key points

  • The study scanned AI 50 companies with a GitHub presence and reports that nearly two-thirds had verified secret leaks.
  • Wiz searched commit histories, fork histories, deleted forks, workflow logs, and gists, as well as public repositories and gists associated with organization members and contributors.
  • The report says companies with identified leaks were valued at over $400 billion in total.
  • Examples included LangChain API keys with organization-level permissions and an ElevenLabs enterprise API key; the article says these leaks were disclosed and promptly fixed.
  • For one company that did not grant disclosure permission, Wiz reports finding a HuggingFace token in a deleted fork that could access about 1,000 private models, along with Weights & Biases keys associated with private-model training data.
  • The report says almost half of disclosures either did not reach their target or received no response, citing missing disclosure channels, non-response, or unresolved issues.
  • Wiz recommends public VCS secret scanning, broader coverage for AI-service credentials and less-visible repositories, and establishing disclosure channels.

Article Details

Publisher
Wiz IO
Scope
GitHub secret exposure scanning of private companies in the Forbes AI 50, excluding companies without a GitHub presence; included organization repositories, commit and fork history, deleted forks, workflow logs, gists, and public repositories of organization members and contributors.
Sample Size
Forbes AI 50 companies, minus those without a GitHub presence; the exact number scanned was not stated.
Key Statistics
  • 65% of the analyzed companies had verified secrets leaks on GitHub.
  • Companies with verified leaks were valued at over $400 billion in total.
  • Almost half of disclosures either failed to reach the target or received no response; the article gives no exact count or denominator.
  • The smallest-footprint company with a verified leak had 0 public repositories and 14 organization members.
  • The largest-footprint company without an exposed secret had 60 public repositories and 28 organization members.
Recommendations
  • Deploy secret scanning for public version-control systems.
  • Establish disclosure channels and processes.
  • Add detection support for proprietary and emerging AI service secret types.
  • Treat organization members and contributors as part of the company’s attack surface; create a VCS member policy, including MFA for personal organizations and separation of personal activity.
  • Continuously update scanning policies to cover new file types and secret vectors.

Vendors

ElevenLabswas similar to the general findings in Part 1, featuring AI-related secrets such as WeightsAndBiases, ElevenLabs and HuggingFace among the most popular impactful secrets. HuggingFace4 - Correlations in related networks like HuggingFace and npmLangChainLess common Tavily, Langchain, NVIDIA-NGC, Cohere, Pinecone, Clarifai, Gemini, AI21 Labs, IBM Watsonx AI, Cerebras, FriendliAI, FireworksAI, TogetherAIWeightsAndBiasesMost common Perplexity, WeightsAndBiases, Groq, NVIDIA API

Products

GitHubsecrets behind. We looked at 50 leading AI companies and found that 65% had leaked verified secrets on GitHub. Think API keys, tokens, and sensitive credentials, often buried deep in deleted forks, gists, andLangSmithLangChain - multiple Langsmith API keys in .py, .ipynb, and .env files, including organization-level enterprise_legacy tier keys with org:manage and org:read permissions to LangChain Inc organization. Beyond the

Related Articles