OpenAI announced a new safety protocol for advanced AI development, emphasizing the need for a national framework to ensure safety. The move comes as the company continues to prioritize rapid progress while ensuring capabilities do not outpace alignment and monitoring.
The company reported that safety protocols will be implemented before training runs, which could result in major capability jumps. This approach is intended to set a standard for the industry, with Altman hoping other companies will adopt similar measures.
The new protocols are built on OpenAI's existing safety infrastructure and target the development of more advanced AI models. These measures will be available starting in the fourth quarter of 2026, initially for enterprise clients and research institutions.
"Pacing does not mean stopping. Progress will continue to be rapid, just not as fast as it could," said Sam Altman, CEO of OpenAI. He added that the company believes the cost of pacing is worth it to avoid recklessness in AI development.
The announcement follows discussions among leading AI companies about self-regulation through an independent oversight body. Altman stressed that the U.S. government's support is crucial for international coordination on AI safety.
OpenAI did not specify the exact timeline for full implementation of the protocols, and the company raised concerns about potential monopolization by larger players. Altman said the industry will continue to innovate while ensuring safety remains a priority.
Source: thedecoder
