OpenAI Chief Scientist Warns of Rapid AI Development, Says 'Extreme Caution' Needed Now

动察 Beating AI Flash: According to Bloomberg, OpenAI Chief Scientist Jakub Pachocki warned that AI is evolving rapidly and becoming increasingly difficult for humans to understand and control. Current AI models can already operate computers, collaborate with humans and other AI systems, and execute research projects. He expects that "recursive self-improvement," where AI enhances itself without human intervention, is not far off.

Pachocki stated: "This is a moment that requires extreme caution. I worry that no one is prepared for the consequences of machine intelligence continuously and rapidly advancing." He believes developers can further steer AI to align with human interests, or slow down development when necessary; until common safety standards are established, proactive deceleration by AI labs should become the norm.

OpenAI previously stated that AI models attacking the Hugging Face research platform coordinated with each other through message boards without being promptly detected. Due to GPT-6 Astra's advanced cybersecurity capabilities, OpenAI chose to restrict its release scope; Anthropic has also not yet made its frontier model Mythos publicly available for similar reasons. Meanwhile, OpenAI still plans to build a "truly automated AI researcher" within two years.