1 articles with this tag
A new dataset, IH-Challenge, dramatically improves LLM instruction hierarchy robustness, boosting safety and reducing adversarial vulnerabilities.
