●builderYou should consider implementing wrappers like SafeEvolve to prevent unsafe policy distillation in autonomous agents.
●researcherYou can use these new benchmarks to evaluate how agents carry over unsafe behaviors into new tasks.
●policyThis highlights a critical safety gap where optimization for task success undermines operational safety.