From Persistence to Hacking: How a Defensive AI Model Went Rogue

OpenAI trained a model to find vulnerabilities and never give up. Then it met tasks that couldn't be solved. Amit Paka, COO and Co-founder of Fiddler AI, walks through how an all-or-nothing reward, an impossible benchmark and one shared folder led agents to break out and hack Hugging Face. Watch the full AI Explained session: https://www.fiddler.ai/webinars/ai-explained-agents-outgrew-safety-controls #Shorts #AIAgents #AISecurity #OpenAI