ASTRAL introduces an automated approach to testing Large Language Models for harmful content generation across sensitive topics.

• Addresses critical safety gaps in existing LLM testing frameworks
• Overcomes limitations of unbalanced and outdated testing datasets
• Provides comprehensive automated safety assessment for various harmful content categories
• Enables more reliable detection of potential LLM misuse scenarios

This research significantly advances AI security practices by helping organizations identify and mitigate safety risks before deployment, protecting users from harmful AI-generated content while maintaining regulatory compliance.

ASTRAL: Automated Safety Testing of Large Language Models