This research introduces a systematic framework to evaluate Custom GPTs against usage policies, addressing critical security and compliance gaps in LLM deployments.

• Enables automated assessment of fine-tuned LLMs against established safety policies
• Identifies potential security vulnerabilities and compliance issues in black-box models
• Provides a structured approach to ensure Custom GPTs adhere to ethical guidelines
• Particularly valuable for evaluating models in the growing GPT marketplace

For security professionals, this framework offers a proactive method to detect and mitigate risks before deployment, ensuring safer integration of custom LLMs into business operations.

Towards Safer Chatbots: A Framework for Policy Compliance Evaluation of Custom GPTs