“I am concerned that no one is prepared for the consequences of the continuous and rapid rise in machine intelligence,” Jakub Pachocki wrote in his blog post.
This post comes just days after the ChatGPT maker launched GPT-6 Astra, which it calls its most powerful product yet.
In addition to OpenAI, other artificial intelligence companies such as Anthropic have made statements that artificial intelligence agents act autonomously and carry out cyber attacks against other companies.
In July, OpenAI said that the incident in which artificial intelligence agents, which are artificial intelligence systems that can work alone with human instructions, hacked the technology platform called Hugging Face, was unprecedented.
Another report claimed that the company's artificial intelligence agents also took over a German website months ago.
'Responses to concerns are not sufficient'
“We are facing a transition to a world of incredibly intelligent machines, and we need to ensure that this transition ends well for humanity,” Pachocki wrote in his latest blog post.
He said OpenAI will continue to "build defenses" and seek technical solutions to adapt. Adaptation is a term used to ensure that a machine's actions and goals match perfectly with human intent and security measures.
Pachocki added that one of the company's priorities in the future will be to develop an "automatic artificial intelligence researcher" that will keep up with developments in artificial intelligence while ensuring that human researchers remain a part of the process.
Professor Gina Neff, Director of the Minderoo Center for Technology and Democracy at the University of Cambridge, evaluated this proposal as follows: "Instead of better AI measures, regulations or safeguards to keep people safe, they propose to develop internal AI agents to investigate these problems."
Such responses to growing concerns that OpenAI's models lead to problems such as cybersecurity, job losses, bugs, inaccuracies and fraud are simply not enough.
Nathan Calvin, general counsel at advocacy group Encode AI, said he agreed with Pachocki about the dangers involved in developing advanced AI models, but argued that OpenAI has failed to be transparent, which means warnings risk being dismissed as “solely self-serving hyperbole.”
“If Jakub and others at OpenAI want relevant people in the AI industry to stick with them to get things going, one of the most important things they can do is share a lot more information about things that make them think they should be cautious,” he wrote in X.
What security measures are available for AI?
Global regulations are having difficulty keeping up with the pace of development of artificial intelligence.
The European Union's Artificial Intelligence Law came into force on August 2 and requires AI giants such as OpenAI to prove that their most powerful models cannot autonomously launch a cyberattack or escape human control before they are allowed to be sold in Europe.
But because the law's jurisdiction is limited to Europe, it cannot prevent an out-of-control AI developed elsewhere from threatening the continent.
In his blog post, Pachocki called for the establishment of legally or internationally mandated minimum security thresholds that could be enforced by a “network of third-party auditors” or “government agencies.”
He cited the need for AI labs to meet these thresholds before they are allowed to continue scaling or rolling out advanced models.
He added that he hopes it will become common for companies to slow down their artificial intelligence development processes until common security measures are established.
In August, OpenAI announced that it was slowing down the training of some of its most advanced artificial intelligence models to improve security.
