AI is surpassing humans on more fronts; OpenAI scientist warns safeguards may not keep pace

Pachocki says chain-of-thought monitoring is becoming less reliable as advanced models learn to manipulate their own reasoning, while cybersecurity and engineered threats raise the stakes. Artificial intelligence labs may soon have to ease off the accelerator, OpenAI Chief Scientist Jakub Pachocki has warned, saying existing safeguards have not reached a point where developers can responsibly keep scaling increasingly powerful systems at maximum speed. Pachocki's warning centres on a widening gap between what advanced AI can do and humanity's ability to understand, monitor and control those capabilities. He said international coordination on future AI development should become a priority for governments, while common safety thresholds must precede unrestricted scaling. Advertisement "This is a time that calls for extreme caution. I am concerned that no one is prepared for the consequences of a continued rapid rise in machine intelligence. OpenAI will continue to seek technical solutions to alignment and monitoring, to build defensive systems and unilaterally withhold further scaling as needed; however, I believe broader interventions are required," Pachocki said. AI systems could increasingly help build more powerful AI One of the concerns raised by Pachocki is the prospect of machine systems taking a larger role in their own development. OpenAI's internal research, he said, suggests the current rate of progress could continue into recursive self-improvement. Such a shift could allow AI systems to contribute increasingly to the creation of more capable successors. Reasoning models are already expanding beyond conventional chatbot functions. They can work with digital interfaces, contribute to scientific research, and operate in areas such as computer security. Pachocki argued that an AI system does not need to outperform a person at everything to have profound real-world consequences. "To become very relevant in the real world - very useful or very dangerous - the…