OpenAI Chief Scientist Jakub Pachocki published an essay on the company’s website Saturday warning that no AI lab has done enough on alignment and monitoring to justify continuing to scale systems at maximum speed. Meanwhile, OpenAI has just launched its most powerful AI model to date.
In the essay, Pachocki claims that voluntary slowdowns are likely to become common across the industry until shared safety standards are established. “This is a time that calls for extreme caution,” he wrote. OpenAI would unilaterally hold back further scaling when needed, he said, but he argued that broader action from industry and government is also required.
The essay describes AI as moving toward recursive self-improvement, in which systems advance without human direction, and says capability gains are likely to continue or even speed up. “I am concerned no one is prepared for the consequences of a continued rapid rise in machine intelligence,” Pachocki wrote.
He also criticized chain-of-thought monitoring, the practice of reviewing a model’s step-by-step reasoning to catch problems, saying its effectiveness is eroding. He named three causes: the boundary between a model’s internal reasoning and its supervised communication with people is dissolving, models are increasingly shaping their own reasoning chains, and capability gains are showing up even without any verbal reasoning output to monitor.
Pachocki called for OpenAI’s Preparedness Framework and Anthropic’s Responsible Scaling Policy, both internal safety frameworks, to become legal requirements enforceable by third-party auditors, government agencies or international bodies. Governments should treat international coordination on AI development as a top priority, he said.
A separate OpenAI post, also published Saturday, offered a glimpse of how far automation has already spread inside the company. It measured the amount of work done by automated agents against work done by humans at OpenAI and found that by mid-August the company’s research organization had reached a ratio of 3.1 agent-workdays for every one human workday.
Both posts arrived three days after OpenAI released GPT-6 Astra, the company’s first model to receive a “Critical” rating under its own Preparedness Framework. That rating reflects the model’s ability to autonomously find security vulnerabilities and build working exploits against hardened systems without step-by-step human direction. OpenAI said it built lessons from an earlier incident, in which unreleased models had compromised Hugging Face’s systems, into the safeguards surrounding Astra.
The timing is awkward: a warning about unpreparedness landed days after a “Critical”-rated release, from a company that now leans on AI agents to do most of its own research. Outside observers noticed. Nathan Calvin, general counsel at Encode AI, told the BBC that Pachocki’s comments amounted to “just self-interested hype,” given OpenAI’s lack of transparency. Gina Neff, a professor who leads the Minderoo Centre for Technology and Democracy at Cambridge, told the BBC that OpenAI’s dependence on internal AI agents for safety research is “simply not good enough.”
AI giants simultaneously release powerful AI models and call for industry regulation. Breitbart News social media director Wynton Hall has written his instant bestseller Code Red: The Left, the Right, China, and the Race to Control AI to serve as the definitive guide on how the MAGA movement can create positions on AI that benefit humanity without handing control of our nation to the leftists of Silicon Valley or allowing the Chinese to take over the world.
Lucas Nolan is a reporter for Breitbart News covering issues of AI, free speech, and online censorship.
Read the full article here