Over 1,100 technology professionals have launched an initiative calling for the establishment of a global governance mechanism to regulate the pace of AI development.

Discussions on AI security are shifting from principle-based declarations to concrete issues regarding how to control the pace of its development. According to a report by Reuters on July 29, more than 1,100 employees from leading technology companies have supported an initiative called “Pacing the Frontier,” which calls on the U.S. government to foster international cooperation in order to develop the technical and regulatory tools needed to manage the speed of advances in cutting-edge AI.

The participants came from organizations such as Meta, Anthropic, OpenAI, and Google; supporting organizations included Guidelight AI Standards and Encode AI. Among those who signed the agreement were Dario Amodei, CEO of Anthropic, several co-founders of that company, Jakub Pachocki, chief scientist at OpenAI, as well as AI research managers from Meta.

The focus of this initiative is not simply to halt innovation, but rather to establish verifiable control mechanisms to address potential rapid recursive improvements. If models can significantly speed up the development of the next generation of models, the pace of industry advancement could shift from monthly to weekly or even daily, making it difficult for traditional legislation, auditing, and corporate risk management practices to keep up.

Effective “rhythm governance” requires more solid tools than mere verbal commitments. For example, threshold levels should be set across institutions for high-risk training tasks, audit records should be kept for computing clusters and the weights of key models, uniform standards should be applied to tests of dangerous capabilities, and verifiable procedures for suspension and notification should be put in place in the event of an incident. Only when different laboratories use compatible standards can international cooperation avoid being slowed down unilaterally.

The initiative also reflects a growing convergence in the assessment of risks within leading laboratories. OpenAI states that the pace of advancement in AI could be so fast that it will be necessary to actively control the development of these advanced models; Anthropic, on the other hand, emphasizes the need to prepare appropriate tools in advance, so that society has time to adapt to potential rapid changes.

For the industry, governance mechanisms do not necessarily undermine competitiveness. Uniform performance evaluations, transparent accident reporting, and trackable safety commitments can reduce corporate clients’ concerns about adopting AI systems with high levels of autonomy. The real challenge lies in preventing standards from being used by a few companies to strengthen their market position, while also ensuring that high-risk capabilities do not spread in the absence of proper regulation.

Source:Reuters