MUMBAI: The AI race is picking up speed, but Mark Zuckerberg says companies do not necessarily need to hit the brakes. The Meta CEO has argued that AI labs already have strong commercial incentives to make safety and alignment a priority as increasingly capable systems move closer to everyday use.
In a post on X, Zuckerberg pushed back against the argument that AI development must be slowed until alignment research catches up with capabilities. He said companies building AI agents have a direct reason to ensure their systems understand users and follow their instructions.
“People won’t want to use agents that are misaligned with them and that don’t do what they ask,” Zuckerberg said, adding that trust and alignment were quickly becoming among the most important capabilities separating AI agents and models.
His comments come as the industry wrestles with a familiar question: how fast is too fast when the technology is advancing faster than researchers can fully understand its risks?
Anthropic CEO Dario Amodei has called for the development of the most advanced AI systems to be paced more carefully. In an essay titled “We Must Pace the Frontier”, Amodei proposed a three-part approach involving independent oversight of AI companies, greater coordination between firms in democratic countries and wider international cooperation.
Amodei clarified that his proposal was not intended to halt AI research or model training. Instead, he argued for giving safety measures more time to keep pace with rapid advances in capabilities.
OpenAI CEO Sam Altman, SpaceXAI CEO Elon Musk and Google DeepMind co-founder and chair Demis Hassabis have endorsed Amodei’s position, according to the information provided.
Zuckerberg, however, argued that individual companies can make their own safety decisions without waiting for the entire industry to move in lockstep.
As an example, Zuckerberg pointed to Meta’s decision to delay the release of its Muse system for several months while the company worked on safety and security.
“Every lab has the responsibility and incentive to move at the pace required to train its models safely, and the ability to take its own actions to ensure that happens,” he said.
Zuckerberg said Meta did not ask other AI labs to take similar action before making its own decision, presenting the delay as an example of an individual company responding to its own assessment of risk.
He also highlighted independent scrutiny as another part of the safety equation. Meta Superintelligence Labs, or MSL, works with independent evaluators and advisers across several areas, Zuckerberg said.
Rather than relying on a small number of evaluators, he argued that the industry should build a larger and more diverse pool of independent experts capable of assessing advanced AI systems.
Zuckerberg also brought computing power into the discussion, saying Meta has committed the “significant majority of compute” towards serving people rather than pursuing recursive self-improvement.
His argument is that how computing resources are allocated can influence the trajectory of AI development, particularly as companies build increasingly capable systems.
The broader issue, he said, is maintaining a balance of power as AI develops.
“I believe the key to building a positive future for everyone is maintaining the right balance of power,” Zuckerberg said.
The disagreement therefore is not simply about whether AI needs safeguards. The emerging fault line is over pace, oversight and who should decide when development has moved quickly enough — or too quickly to warrant a pause.