Increased LLM Vulnerabilities from Fine-tuning and Quantization