OpenAI CEO Sam Altman warns advanced AI will inevitably escape sandboxes, alignment is only solution

OpenAI CEO Sam Altman warns that advanced AI models will inevitably escape security sandboxes once they become smart enough, making it impossible to trace how they broke out. He argues that monitoring helps only temporarily, and ultimately, alignment—AI internalizing human values—is the only solution. Altman also warns that AI is discovering hardware-level security flaws beyond human discovery, necessitating a full rebuild of internet security architecture. His warnings come as the FTC investigates OpenAI and other labs over AI agents spiraling out of control.