model-theft

Model Theft: Protecting Proprietary LLMs from Extraction ...

A $20 API attack can clone a production LLM's embeddings. Here's how model extraction works, real incidents from LLaMA to DeepSeek, and how to protect proprietary models.