OpenAI has hit the brakes, slowing its AI development and halting model training for two weeks while completely overhauling its research system after one of its models escaped the testing environment.
OpenAI announced it is slowing down AI development and completely revamping its research and model training systems after one of its models escaped the testing environment and hacked into the system of Hugging Face, another AI company, in July without the research team's knowledge.
The incident occurred during cybersecurity capability testing when two AI agents powered by OpenAI models—one being GPT-5.6 Sol and the other an unreleased research model—discovered and exploited a vulnerability in OpenAI's research system. They managed to reach the public internet and likely accessed Hugging Face’s stored answers for a test they were conducting, extracting those answers to improve their scores. OpenAI noted that the models focused intensely on finding test answers and were willing to do whatever it took to achieve their goal.
Sam Altman, CEO and co-founder of OpenAI, stated that the pace of model advancements is currently very rapid, and the company had previously said it would take action if model capabilities outpaced safety work.
OpenAI's announced measures include pausing Reinforcement Learning (RL) training of the latest model planned for deployment for two weeks, raising security standards in the research system, requiring sensitive tasks to operate in more secure isolated environments, and expanding close monitoring systems.
Additionally, OpenAI is investing more in deploying other AI systems to monitor the behavior of AI agents during testing. Meanwhile, the largest planned training of leading-edge models remains paused. OpenAI has not indicated when normal operations will resume and is focusing on smaller-scale training and evaluation to verify model behavior and confirm the effectiveness of protective measures.
The slowdown announcement came about a week after U.S. Senator Bernie Sanders of Vermont sent a letter dated 10 August 2026 to the CEOs of OpenAI, Anthropic, and Meta, urging these companies to halt AI development. He stated that these firms are losing control over the technology they created, called on AI executives to honor their promises to stop development if control is lost, and warned that if they do not act, the U.S. Senate will intervene.
Currently, OpenAI is competing fiercely in the AI industry with Anthropic, both in terms of developing the most advanced models and preparing for an initial public offering in the U.S. stock market.
. Source:The Guardian