The debate over how quickly the world should develop increasingly powerful artificial intelligence is gaining momentum, with some of the biggest names in the industry now signaling support for greater oversight and a more measured approach to frontier AI development.
Anthropic CEO Dario Amodei has called for policymakers and AI companies to consider ways to “pace” the development of frontier systems, particularly as models move closer to artificial general intelligence (AGI). His position has received notable support from OpenAI CEO Sam Altman and Elon Musk, putting executives from several leading AI companies on broadly similar ground over the need for stronger safety measures.
Musk publicly backed Amodei’s position, while Altman also expressed support for independent evaluation and greater transparency around the development of advanced AI systems. The alignment is notable given the long-standing disagreements between some of these industry leaders.
Independent Evaluators Proposed
One of the key ideas gaining attention is the use of independent, third-party evaluators embedded within frontier AI companies.
Under the proposal, organizations focused on AI safety could have significant access to model-training processes and systems, allowing them to independently assess how advanced models are being developed and whether appropriate safeguards are in place.
The objective would be to create an additional layer of accountability without requiring companies to expose sensitive customer information or proprietary technology.
Supporters argue that independent oversight could reduce the risk of AI companies prioritizing speed and competition over safety. Critics, however, could question how truly independent such evaluators would remain when operating inside companies they are supposed to monitor.
The Competition Problem
A major challenge with slowing frontier AI development is competition.
If companies such as OpenAI and Anthropic deliberately reduce the pace of their most advanced research, competitors including Meta, Google, Mistral and Chinese AI developers could potentially close the gap.
That creates a difficult policy question: Can the leading AI companies slow down without simply handing their technological advantage to competitors?
The issue becomes even more complicated at the international level. A slowdown among U.S. companies could potentially give Chinese AI developers additional time to catch up, raising concerns about whether unilateral restrictions would actually improve global AI safety.
Meta CEO Mark Zuckerberg has previously emphasized accelerating AI development and making powerful AI broadly available. His response to the latest calls for frontier pacing could therefore be closely watched.
Open-Source AI Complicates Safety Efforts
The rise of increasingly capable open-source models adds another layer of difficulty.
Even if the leading proprietary AI companies introduce strict safeguards, those measures become less effective if an open-source model reaches comparable capabilities without implementing the same restrictions.
This creates a fundamental tension between AI safety and open access. Frontier companies may be able to impose safeguards on their own systems, but controlling how independently developed models are trained, modified and deployed is considerably harder.
The gap between proprietary and open-source AI therefore remains an important factor in any attempt to regulate frontier development.
The Rise of Local AI
At the same time, advances in open models are making it increasingly realistic for individuals and small organizations to operate powerful AI systems locally.
Models such as DeepSeek have demonstrated how capable AI can increasingly be deployed outside traditional hyperscale data centers. Falling hardware costs and improving model efficiency could eventually make private AI laboratories accessible to a much wider audience.
Local AI could offer an important privacy advantage because sensitive information can potentially remain on a user’s own hardware rather than being processed by a remote AI provider.
However, running powerful models locally also creates a regulatory challenge: the more capable AI becomes on consumer hardware, the harder it becomes for governments and companies to control who can access advanced intelligence.
The Bigger Question: Do We Need AGI Now?
The discussion ultimately raises a broader question about the purpose of AI development.
Some argue that humanity can obtain enormous benefits from specialized AI systems without immediately pursuing AGI. AI systems designed for research, programming, business automation, scientific analysis and other specific tasks could continue improving while presenting potentially fewer risks than a highly autonomous general intelligence.
Others believe that continued progress toward increasingly general systems is inevitable and could unlock breakthroughs in medicine, science, energy and other fields.
The disagreement is therefore not simply about whether AI should be developed faster or slower. It is about which capabilities should be developed, under what safeguards and who should ultimately control them.
A New Phase in the AI Race
The growing support for independent evaluators and frontier AI pacing suggests that the industry’s safety debate may be entering a new phase.
The immediate proposal is not necessarily a complete halt to AI development. Instead, the emerging argument is for greater transparency, independent evaluation and potentially rules that apply across the entire frontier AI industry rather than relying solely on voluntary commitments.
Whether governments ultimately impose such requirements remains uncertain. But with major AI leaders increasingly discussing the need for oversight, the question of how fast frontier AI should advance is moving beyond a niche safety debate and toward the center of the global technology conversation.