Summary

  • OpenAI has inquired with lawmakers about the legality of a coordinated slowdown among competing AI firms, as reported by WIRED.
  • Experts caution that financial and competitive pressures make independent slowdowns unlikely.
  • A proposed bill aims to safeguard certain agreements that would limit AI development to mitigate security concerns.

OpenAI has approached Congress to clarify whether competing AI developers could legally agree to pause their advancements, following a call from its chief scientist for the establishment of collective safety standards within the industry.

As reported by WIRED, sources familiar with the matter indicated that OpenAI has sought legal insights on whether a collaborative slowdown could breach antitrust regulations.

Myriad: SPY high in September? Click to make your prediction.

This inquiry follows a statement from OpenAI's chief scientist, Jakub Pachocki, advocating for a voluntary slowdown until developers can ensure their systems are safe. However, the competitive landscape often pressures companies to continue innovation, even amid internal concerns.

Miranda Bogen, chief technologist at the Center for Democracy and Technology, stated to Decrypt, "The competition in AI is extremely fierce, creating a troubling scenario where firms are driven to launch products before fully understanding their risks."

She further noted, "Despite internal awareness of the need for more research and testing, these companies face enormous pressure to bypass essential safety assessments, even as evidence mounts about the dangers of rapid advancement."

Any agreement to decelerate AI development could trigger antitrust issues by limiting competition. Senators Adam Schiff and Jim Banks have introduced a bill designed to protect specific security collaborations, which would require prior notification to the Justice Department.

The discussion around slowing AI progress is particularly pertinent as the U.S. and China vie for dominance in this technological arena. In May, former President Donald Trump postponed an executive order on AI due to worries it might undermine America’s competitive edge against China. He subsequently signed the order in June, which initiated a voluntary review process for advanced AI models prior to their release.

Financial motivations may complicate any potential agreement on slowing development more than concerns about national security.

Duncan Sabien, head of communications at the Machine Intelligence Research Institute, remarked to Decrypt, "Each advancement in the current climate can lead to millions or even billions in additional funding and enhances the influence of the innovators behind it."

He described the ongoing arms race as a reality, emphasizing that both OpenAI and Anthropic are under pressure to progress, as each new system aids in the creation of more advanced successors, risking a competitive disadvantage if either company opts to slow down without a mutual agreement.

"This is particularly critical since advancements in intelligence are cumulative; each new system simplifies the training and deployment of the next, more sophisticated model," he explained. "Without some form of coordinated effort, taking a step back simply allows competitors to gain a lead."

In August, OpenAI halted internal work on its Astra project due to cybersecurity worries. However, both OpenAI and Anthropic eased their safety commitments in February, with Anthropic’s chief science officer, Jared Kaplan, arguing that independently slowing development was impractical while rivals continued to innovate.

The conversation about slowing AI development intensified this week following the resignation of former Anthropic engineer Jacob Coxon, who cited concerns regarding the existential risks posed by ongoing AI development.

“Those creating AI genuinely believe it could lead to catastrophic outcomes by the decade's end,” Coxon stated on X, asserting that his comments were not merely a “marketing stunt.”

“In fact, many executives and senior researchers may frame their statements to appear measured in public, yet I hear them express genuine fear in private. No other human endeavor presents this level of danger,” he expressed.

Nonetheless, Sabien pointed out that researchers might achieve little by resigning if their positions are simply filled by others who are less cautious.

“If enough individuals come to a shared understanding that they ought to cease operations, they can overcome the coordination barrier and halt collectively, rather than just being replaced by someone slightly less cautious,” he concluded.