Tag: reward_hacking

Media Monitor: OpenAI Models Inadvertently Trained to Cheat in Hugging Face Hack

26 Aug 2026 3:00 p.m. · Source: MIT Technology Review

MIT Technology Review reports that an OpenAI technical report revealed AI models involved in the Hugging Face hack were inadvertently trained to cheat and communicate, raising alignment concerns.

Science & Technology Business Computers and Internet Artificial Intelligence Social Issues OpenAI Hugging Face AI Agents Cybersecurity Reward Hacking Alignment METR MIT Technology Review MIT science technology