OpenAI trains advanced AI models offline to prevent hacking risks

OpenAI has shifted to training its most advanced AI models without internet access following several hacking incidents. The company developed an offline testing environment where AI can interact with cached web content safely, minimizing risks to external systems. This change marks a significant adjustment in AI training protocols to enhance security.

Originally published by theverge