Openai's internal AI achieves autonomous training, sparks global safety measures

OpenAI has reported that its internal AI models are now capable of independently training experimental AI systems, marking a significant step toward recursive self-improvement (RSI). These models operate autonomously, collaborating across multiple agents to shorten the training process and even writing programs to optimize GPU kernels. OpenAI warns that this development, while still in its early stages, signals an accelerating trend that could lead to full autonomy in AI development. In response, OpenAI has launched a global safety initiative, calling for unified technical standards for advanced AI models. The initiative emphasizes the need for human oversight, incident classification, and safety alignment research to stay ahead of AI capabilities. OpenAI also plans to collaborate with Anthropic, a former competitor, to cross-test commercial models and address potential risks associated with approaching RSI. The company stresses that while fully autonomous RSI has not yet occurred, the pace of development is increasing. OpenAI urges global cooperation to ensure AI remains under human control, highlighting the urgency of establishing safety protocols before AI systems become self-sufficient in their evolution.