Frontier AI slowdown debate sharpens as Sacks backs voluntary pacing, Meta prioritises alignment
David Sacks says OpenAI and Anthropic should slow frontier AI voluntarily if risks warrant it, as Meta's Alexandr Wang flags alignment as key to scaling.
A debate over whether to slow the development of frontier artificial intelligence has gathered pace after Anthropic CEO Dario Amodei called for "pacing the frontier" and OpenAI CEO Sam Altman endorsed the idea.
Responding to those calls, David Sacks, Chair of the President's Council of Advisors on Science and Technology and a former White House AI and Crypto Czar, said the two companies do not require broader regulatory restrictions to slow their own work. In a post on X, he urged them to go ahead if they believe their unreleased models carry risks serious enough to justify a pause, noting that both firms sit at the frontier of AI development.
Sacks also questioned whether safety concerns were the sole motive behind slowdown proposals. He pointed to potential product-liability exposure should AI models be used to enable a damaging cyberattack, and said the market already penalises models that behave unpredictably or without authorisation. Trading some raw capability for reliability and predictability, he argued, is simply good business and gives customers what they want.
On regulation, Sacks said pacing the frontier could create room for a more considered discussion than blanket calls to shut the technology down. Any global approach, he added, would have to account for China, which he said is unlikely to join an international agreement.
He further warned that if companies are unwilling to slow down voluntarily, pressing for their preferred regulatory framework could be seen as an attempt to protect their competitive position or shape the political debate.
Separately, Meta Chief AI Officer Alexandr Wang said alignment is becoming increasingly important as models grow more capable. In a post on X, he described alignment as fundamental to delivering personal superintelligence, stressing that people need agents they can trust to do reliably what they are asked.
Wang said Meta Superintelligence Labs is rapidly increasing the share of its effort devoted to alignment as its models become more powerful, and that alignment could prove the gating factor for scaling closer to the frontier.