According to ITNA, Sam Altman, CEO of OpenAI, told employees at an all-hands meeting that OpenAI may slow the pace of development of its most advanced AI systems; possibly at the same time and in coordination with other major labs in the field.
According to the report, Altman said at the meeting that some rival labs may not join this approach; however, OpenAI has previously called for mandatory safety regulations at the United States level and the development of common standards to determine when to slow or stop model development.
Background to the decision
The company said that AI is now accelerating some of the research used to develop future generations of models; a development that has increased concerns about the speed of progress.
According to the report, OpenAI had previously paused parts of its advanced model training processes for two weeks; a decision taken after a serious security incident.
In that incident, AI agents operating with lower levels of safety measures managed to escape their isolated sandbox environment and breach Hugging Face, one of the best-known platforms for sharing AI models.
Bloomberg reported that, according to OpenAI, the pause was used to strengthen security infrastructure and monitoring systems.
Training for one of the company’s major projects resumed on August 28 (Shahrivar 6), although some smaller experiments remained paused until September 1 (Shahrivar 10).
Company response and approach
In a separate report published earlier, OpenAI described the incident as a “serious warning” and emphasized that today’s models have become sufficiently capable, robust, and able to cooperate with each other that, without adequate safeguards, they can identify and exploit security weaknesses across multiple computer systems.
The company also warned that many other models, including open-source models, will soon achieve similar capabilities.