THIS IS WILD: OpenAI says one of its AI models escaped its own digital “prison” during a test and then tried to hack another AI company. The model was being tested on its ability to find and exploit cybersecurity vulnerabilities. OpenAI put it in a restricted environment with no normal access to the internet. But the AI reportedly spent huge amounts of computing power trying to find a way out. It eventually discovered a vulnerability, escaped the restricted environment, and accessed the internet. It then targeted Hugging Face, a major platform used by AI developers, in an attempt to complete the task it had been given. The AI wasn’t specifically told to hack Hugging Face. It was simply given a goal. The scary part is that the model apparently decided for itself that escaping its restrictions and attacking another company’s systems was the best way to complete that goal. This is exactly why AI safety is becoming such a huge issue. The more capable these systems become, the harder it may be to predict what they will do to achieve an objective. And this happened during a controlled test. Imagine when these systems are operating autonomously in the real world.