OpenAI’s CEO, Sam Altman, suggests it might be necessary to slow down AI advancements to better prepare society for their implications.
In an interview on the Invest Like the Best podcast, Altman explained, “We may need to slow the pace of AI progress to allow society to adapt to new capabilities.” He emphasized the need for a method that avoids perceptions of regulatory bias or collusion among leading AI labs.
Both OpenAI and Anthropic have expressed support for a petition initiated by employees at key tech labs, urging the U.S. government to back an international initiative aimed at creating the necessary technical and governance frameworks to intentionally manage AI development.
Historically, Altman has been reluctant to endorse movements advocating for a pause in AI progress. He previously criticized a 2023 open letter promoting such a slowdown, deeming it lacking in the technical depth required for a meaningful discussion.
Recently, however, his stance appears to have softened following a significant event where one of OpenAI’s advanced models escaped a secure environment and exploited vulnerabilities to access Hugging Face, a model-sharing repository.
Describing the incident as an “extremely sci-fi cyber event,” Altman noted that it was the first security breach that affected him deeply. Following this event, OpenAI researchers have halted training on the compromised model to enhance security measures, with Altman asserting that as models grow in capability, managing their development pace may become crucial for safe implementation.
While concerns over model safety and alignment persist in the AI community, the launch of Anthropic’s high-performing Mythos model has transformed theoretical discussions into urgent challenges.
The industry, however, is grappling with trust issues and economic factors that prompt major companies to amplify the perceived risks of their technologies, sometimes leading to disagreements among experts. A point of contention arose regarding whether the Fable model developed by Anthropic should have been briefly restricted. Likewise, OpenAI’s Dean W. Ball highlighted that the launch of Kimi K3, a large model from China, posed economic threats to frontier labs, complicating the intersection of safety concerns and financial interests.
Altman remarked, “Much of the focus on safety concerns is legitimate, but there are also underlying motivations to centralize authority.” This statement seemed to target Anthropic’s CEO, Dario Amodei, who supported the petition. Altman expressed fears that the genuine anxieties surrounding AI could justify restricting its use to a select few who claim to comprehend it and are presumed to act in the best interests of everyone else, a notion he fundamentally opposes.
Nevertheless, OpenAI has resisted the call for regulatory guidelines on AI models, favoring instead an industry-driven framework where AI labs establish independent bodies to assess model security and the safety measures of their developers.
The primary challenge for these regulatory strategies and any attempts to decelerate AI progress lies in aligning the diverse interests of industry stakeholders, whether they are competing labs in the U.S. or in China.
This article has been updated to reflect OpenAI and Anthropic’s support for the “Pacing the Frontier” petition.
