Safety Alignment Broken During Fine-Tuning

When an organization outsources model fine-tuning without verifying the contents of the fine-tuning dataset; due to unmitigated control gaps, after deployment, the model produces content that should have been blocked under specific inducement, and only then does the organization discover that safety alignment was compromised during fine-tuning, triggering compliance exposure and operational reputational costs.