OpenAI is ready to slow down the race for the most powerful AI — but it doesn’t want to do it alone. Sam Altman told employees that the company could slow down the pace of developing advanced models if other major AI labs joined in.
OpenAI is ready to slow down the race for the most powerful AI — but it doesn’t want to do it alone. Sam Altman told employees that the company could slow down the pace of developing advanced models if other major AI labs joined in.
This is reported by Bloomberg, citing people familiar with the proceedings of an internal OpenAI meeting this week.
According to them, Altman allows for a coordinated slowdown in development along with several competitors, although he admits that not everyone will agree to this.
Also this week, OpenAI officially called for common rules for the most powerful AI models. The company believes that developers and states should agree in advance on the conditions under which AI development should be slowed or stopped if the systems become too dangerous.
And such an agreement would have to be legal. According to WIRED, OpenAI has been reaching out to U.S. lawmakers in recent weeks to see if a joint slowdown in AI development would violate U.S. antitrust laws. And when several major competitors agree to limit the scope of their activities, it could potentially be viewed as anticompetitive coordination.
So one idea is to get the government involved. In July, more than 1,300 employees of companies developing advanced AI signed a petition called Pacing the Frontier. They asked the US government to help create an international mechanism that would allow the entire industry to simultaneously slow down the development of the most powerful systems, if necessary.
Talk of a slowdown has intensified after several incidents with autonomous AI agents and the emergence of significantly stronger models.
The new GPT-6 Astra is the first OpenAI model to be given a critical cyber capability rating by the company. OpenAI estimates that Astra is already capable of independently finding previously unknown vulnerabilities in well-protected systems and developing ways to exploit them. Because of this, OpenAI has previously delayed certain stages of Astra’s development to strengthen its defenses and test safeguards before release.
Previously, dev.ua wrote that Anthropic suggested that competitors agree to temporarily slow down the training of superpower AI due to the risk that the systems would increasingly automate their own development.
