GMP: A Benchmark for Content Moderation under Co-occurring Violations and Dynamic Rules

Researchers introduce GMP, a new benchmark highlighting critical challenges in AI content moderation systems when dealing with co-occurring policy violations and dynamic platform rules. The study reveals that current large language models struggle with consistent moderation when policies are unstable or context-dependent, leading to either over-censorship or allowing harmful content.

• →AI content moderation systems face significant challenges with co-occurring violations where single posts violate multiple policies simultaneously.
• →Dynamic and context-dependent moderation rules expose core limitations in current large language model judgment capabilities.
• →High performance on static benchmarks does not guarantee robust AI generalization to real-world moderation scenarios.
• →Current AI shortcomings lead to inconsistent moderation decisions, creating risks of censoring legitimate content or missing harmful posts.
• →The GMP benchmark addresses a critical gap in evaluating AI systems for real-world content moderation applications.