12,000+ API Keys and Passwords Found in Public Datasets Used for LLM Training
Hard-coded credentials in datasets pose severe security risks for users and organizations.
Large language models may amplify insecure coding practices due to the presence of live secrets in training data.