OneBench
Sharding Prevents LLM Oversight Failures and Adversarial Exploitation | OneBench: AI Insights