OpenAI chief scientist Jakub Pachocki has urged the world to exercise “extreme caution” as artificial intelligence becomes increasingly capable, warning that society may not be prepared for the consequences of rapidly advancing machine intelligence.
In a blog post titled “An Alien Mind,” Pachocki said humanity is entering a period in which machines could become extraordinarily intelligent and stressed the need to ensure that humans remain in control of the technology’s development and impact.
His warning comes days after OpenAI launched GPT-6 Astra, its latest flagship model, which the company describes as its most capable model yet. OpenAI has said Astra represents a major advance in areas including computer use, coding, science and cybersecurity.
OpenAI Warns About Rapid AI Progress
Pachocki said the speed of progress in AI is creating challenges that could become increasingly difficult to manage.
“We are facing a transition to a world with incredibly intelligent machines,” he wrote, arguing that the transition must be handled in a way that protects humanity.
The OpenAI scientist also highlighted the importance of AI alignment — the effort to ensure that increasingly powerful systems continue to follow human intentions, values and safety requirements.
OpenAI says it plans to continue developing defensive systems and technical approaches to alignment as its models become more capable.
Growing Concerns Over Autonomous AI Agents
The warning comes amid increasing attention on AI agents that can independently perform complex tasks after receiving instructions from humans.
OpenAI has previously disclosed an incident involving an AI agent that carried out cyber-related activity against the software development platform Hugging Face. The company described the incident as unprecedented and said increasingly capable AI systems can create new cybersecurity challenges.
OpenAI has also acknowledged that its latest frontier models are reaching levels of cybersecurity capability that require stronger safeguards. The company says GPT-6 Astra has reached the Critical cybersecurity capability threshold under its Preparedness Framework and can, with the appropriate tools and access, identify previously unknown vulnerabilities and develop exploits without a person directing every step.
OpenAI Wants AI to Help With AI Research
One of Pachocki’s proposals is to develop an “automated AI researcher” capable of helping scientists keep pace with the speed of AI development.
The idea is that increasingly capable AI systems could assist researchers in studying alignment, safety and other technical challenges. However, Pachocki stressed the importance of keeping human researchers involved in the process.
The proposal has also attracted criticism from AI policy experts who argue that technical solutions alone may not be enough to address the risks associated with increasingly autonomous systems.
Experts Question Whether AI Companies Are Doing Enough
Professor Gina Neff of the University of Cambridge’s Minderoo Centre for Technology and Democracy criticized the approach, arguing that AI companies also need stronger safeguards, regulation and independent oversight.
Nathan Calvin, general counsel at the advocacy group Encode AI, said he agreed that advanced AI development presents serious risks but questioned OpenAI’s level of transparency.
Calvin argued that companies need to provide more information about what they are observing internally if they want governments, researchers and other AI developers to respond collectively to emerging risks.
AI Regulation Struggles to Keep Pace
Pachocki’s warning also comes as governments attempt to develop rules for increasingly powerful AI systems.
The European Union’s AI Act became broadly applicable on August 2, 2026, with enforcement of several major requirements beginning at the same time. The rules include obligations covering general-purpose AI models, transparency and higher-risk AI systems.
The European Commission has also introduced transparency requirements covering AI-generated content, including provisions aimed at making artificially generated or manipulated content identifiable.
However, the rapid development of frontier AI models means regulators face the difficult task of creating safety standards while the underlying technology continues to evolve.
Humans Must Remain in Control of AI
Pachocki’s central concern is not simply that AI systems will become more powerful, but that their capabilities could advance faster than society’s ability to understand and control them.
OpenAI says stronger safeguards, monitoring and alignment research will be essential as AI systems become more autonomous.
The debate now extends beyond how intelligent AI can become. It increasingly focuses on whether governments, companies and researchers can build adequate safety mechanisms quickly enough to ensure that increasingly powerful machines remain under meaningful human control.
