OpenAI chief scientist warns advanced AI could manipulate, bargain with or blackmail humans: ‘No lab has solved…'

OpenAI Chief Scientist Jakub Pachocki has issued a stark warning about the pace of artificial intelligence development, saying humanity is moving towards a future of increasingly capable machines without yet having adequate safeguards to ensure they remain under human control.

In a blog post titled An Alien Mind, published on September 6, Pachocki reflected on the rapid progress of reasoning models and the prospect of machines becoming more capable than humans in a growing range of tasks. He argued that the immediate focus should be on ensuring that the transition to increasingly powerful AI systems does not come at the expense of human agency.

Pachocki wrote that he believes “no lab has solved alignment and monitoring to a sufficient degree to continue responsibly scaling at maximum speed for much longer.” He also said he expects and hopes that voluntary slowdowns will become more common until shared safety standards are established.

OpenAI scientist warns of a new phase of AI development

Pachocki's concerns extend beyond the effect of AI on jobs and productivity. He described a future in which advanced systems could operate computers, collaborate with people and other AI systems, and carry out research projects with increasing levels of autonomy.

He said the challenge is not simply developing more capable machines but ensuring that humans remain able to direct and control the technology as its capabilities expand.

“We are facing a transition to a world with incredibly intelligent machines,” Pachocki wrote, arguing that preserving human agency should be a central priority as AI becomes capable of performing more tasks.

He also warned about the possibility of an “alien intellect” exceeding human capabilities and said unchecked technological progress could leave people behind.

‘No lab has solved alignment’

AI alignment refers broadly to the challenge of ensuring that increasingly capable AI systems behave in ways consistent with human intentions and values. Monitoring is another major concern as models become capable of carrying out longer and more complex sequences of actions.

Pachocki said existing approaches are not yet sufficient to support an indefinite continuation of AI development at maximum speed.

“Currently I believe that no lab has solved alignment and monitoring to a sufficient degree to continue responsibly scaling at maximum speed for much longer,” he wrote.

His comments come as AI companies continue to push the capabilities of frontier models. OpenAI introduced GPT-6 Astra on September 3, describing it as its most capable model and highlighting advances in computer use, browsing, software engineering, cybersecurity, science and professional work.

OpenAI’s latest model highlights safety challenge

OpenAI's own safety assessment for GPT-6 Astra says the model represents a significant increase in cybersecurity capabilities and that the company strengthened safeguards against harmful actions arising from either misuse or misalignment. The company also said it introduced additional monitoring of model behaviour.

That development provides context for Pachocki's argument that AI progress needs to be accompanied by stronger safety mechanisms. His concern is not that increasingly capable AI is inherently harmful, but that technological progress could move faster than society's ability to manage its consequences.

Call for international coordination

Pachocki argued that the challenge cannot be addressed by AI laboratories alone. He called for voluntary slowdowns while common safety standards are developed and said international coordination on future AI development should become a priority for governments around the world.

He also warned about the risk of an extreme concentration of power if technologies capable of performing work that once required thousands of experts become accessible to only a small number of people operating powerful computers.

For Pachocki, the central question is therefore not simply how intelligent AI systems can become, but whether humans can retain meaningful control as those systems grow more capable.

“We need to ensure that humans remain in control of the future and are not left behind by unchecked progress,” he wrote.