OpenAI Finds Agents That Breached Hugging Face Were ‘Reward Hacking’

OpenAI released a report on the Hugging Face noting that its AI agents are prone to reward hacking, and gaming cybersecurity evaluations.