Fine-tuning and training large language models (LLMs) on sensitive enterprise data brings significant risk if not handled properly. This session will cover the security and privacy concerns that come with LLM training -- from data provenance to leakage risk -- and how to mitigate them. The speaker will share actionable best practices and tooling to ensure enterprise LLM projects remain compliant and safe. Speakers with experience in model development, MLOps or data protection are especially welcome. Potential topics include:

• Safeguards for using PII or proprietary data
• Isolated training environments and access control
• Red-teaming and post-training model testing
• LLMOps and CI/CD pipeline security