Llama-Guard-3-8B

anchor

Findings

• IC-242Most LLMs exhibit higher bias ratios in multi-turn dialogues than in single-turn, with bias accumulating across successive turns [eval]
• IC-244No LLM demonstrates consistently strong fairness across both comprehension-focused and bias-resistance multi-turn tasks; models show complementary failure patterns [eval]