OpenAI Chief Scientist Jakub Pachocki has called for greater caution over the pace of artificial intelligence development, warning that current safety and monitoring techniques may not be sufficient if machine intelligence continues advancing rapidly.
In an essay titled “An Alien Mind,” published on September 6, Pachocki said he was concerned that society was not adequately prepared for the consequences of increasingly capable AI systems. His comments come days after OpenAI released GPT-6 Astra, its latest flagship model, which the company has described as its most capable model to date.
Pachocki said reasoning models have progressed from generating responses to operating computers, collaborating with people and other AI systems, conducting research and playing a growing role in cybersecurity. Based on OpenAI’s internal research, he said the current rate of progress could eventually extend to recursive self-improvement, where AI systems increasingly contribute to improving their own capabilities.
The OpenAI scientist argued that this possibility requires stronger safeguards around how increasingly powerful models are developed and deployed. He said OpenAI would continue working on alignment, monitoring and defensive technologies, while being prepared to withhold further scaling when necessary.
Pachocki also said he does not believe any AI laboratory has yet solved alignment and monitoring well enough to continue scaling frontier models at maximum speed indefinitely. He expects voluntary slowdowns among AI developers to become more common until shared safety standards are established.
His warning comes as AI companies increasingly develop agents capable of carrying out multi-step tasks with limited human intervention. These capabilities have also raised cybersecurity concerns, particularly as models become better at identifying vulnerabilities, writing code and interacting with external computer systems.
Pachocki's comments follow the release of GPT-6 Astra, which OpenAI has classified at the “Critical” level for cybersecurity capability under its Preparedness Framework. The designation reflects the model's advanced abilities in areas including vulnerability discovery and exploitation, while OpenAI has said safeguards are being used to restrict potentially harmful applications.
Beyond technical safeguards, Pachocki called for international coordination on advanced AI development to become a priority for governments. He argued that future systems could create challenges around concentration of power, human agency and control over increasingly automated economic and technological activity.
The remarks add to a broader debate within the AI industry over how quickly frontier models should advance as their ability to act autonomously and perform complex tasks continues to increase.
Disclaimer: This article may include information derived from interviews, press releases, public statements, research, company communications and other publicly available or third-party sources. Such material may be summarised, paraphrased or contextualised for journalistic and editorial purposes. All rights in third-party content remain with their respective owners.