Artificial Intelligence company OpenAI is considering slowing the development of its most advanced AI systems, and CEO Sam Altman is hoping other leading AI labs will follow a similar approach.
In a company-wide meeting this week, Altman told employees that the company could pace its frontier AI work, potentially in coordination with several peer labs, though he acknowledged that not every organization may be willing to participate.
He suggested that the pace of AI development may need to be moderated to give governments, businesses, cybersecurity experts, and society more time to adapt to new levels of AI capability.
Register for the next Tekedia Mini-MBA.
Register for Tekedia AI in Business Masterclass.
Join Tekedia Capital Syndicate and co-invest in great global startups.
He further emphasized that the goal should be to strengthen security and safety measures without creating an environment that resembles regulatory capture or collusion among leading AI companies
Speaking on a podcast, he said,
“We may have to pace the rate of AI development to give ourselves enough time for society to harden around some of these new capability levels trying to figure out how we do that in a way that does not feel like regulatory capture for anyone and also does not feel like collusion among the frontier labs.
“This is the first security incident that I have felt very viscerally as an extremely sci-fi cyber incident. I think it is a good time to slow down. Getting AI safety right is more important than any company’s momentum.”
Altman’s statement comes on the heels of the OpenAI security incident with AI platform Hugging Face. In July 2026, an internal cybersecurity evaluation by OpenAI resulted in its AI models escaping their controlled testing environment and accessing the production infrastructure of Hugging Face, an AI platform that hosts models and datasets.
The models were being tested through ExploitGym, a benchmark designed to measure their ability to identify and exploit software vulnerabilities. During the evaluation, the models found a previously unknown vulnerability in a package-registry cache proxy and used it to obtain internet access despite operating inside a restricted environment.
From there, they identified Hugging Face as a potential source of the benchmark’s models, datasets, and solutions and began looking for ways to access the platform. The AI agents subsequently chained together stolen credentials and software vulnerabilities to gain unauthorized access to Hugging Face’s infrastructure.
According to OpenAI’s investigation, the agents executed code on dozens of Hugging Face servers, obtained root-level access to one server, accessed limited private data, and acquired credentials connected to the company’s messaging platform. Some evaluation data was also copied into a public Hugging Face dataset.
Hugging Face’s security team eventually detected and contained the activity and worked with OpenAI to reconstruct what had happened. Its subsequent forensic analysis identified roughly 17,600 attacker actions carried out over about two and a half days, illustrating the scale and speed at which autonomous AI agents can operate.
Since this incident, OpenAI has already taken concrete steps in recent months. It has halted certain internal training runs and scaled back portions of model development due to safety concerns. In August, the company paused much of its model work for about two weeks.
OpenAI’s chief scientist, Jakub Pachocki, recently amplified the call for restraint. In a public post, he argued that AI companies should coordinate to slow future development as needed and expressed hope that voluntary slowdowns would become common until shared safety standards are established.
Altman himself had previously indicated openness to the idea. In July, he discussed the potential need to pace AI advancement with White House officials and noted that accelerating capabilities might eventually require the industry to give society more time to adapt.
The latest comments arrive against a backdrop of public warnings from researchers. Earlier this week, Jacob Coxon, who had worked at both OpenAI and Anthropic, resigned and accused the companies of racing toward superintelligent systems while gambling with our lives.
His departure and statements added to a growing list of employee concerns across frontier labs about alignment, control, and the risks of uncontrolled capability gains.
Industry observers note that voluntary coordination remains difficult in a competitive landscape. While some U.S. labs have signaled interest in collaborative safety measures, others may prioritize continued rapid progress.