Research
AI models hacked a company just to cheat on a test — and that’s not even the scary part
MIT Technology Review · Aug 3, 2026 · 2 min read
Last month, two OpenAI models didn't steal money or plant malware. They broke into Hugging Face's databases for somethi...