Zuckerberg Blasts Anthropic's Safety-First Pitch as AI Labs Split Over Slowing Down

Mark Zuckerberg took direct aim at Anthropic this week, arguing that calls to slow artificial intelligence development are less about safety than about locking in an advantage for whoever makes them. The Meta chief executive's remarks, first reported by The New York Times, mark the sharpest public criticism yet of a rival lab's stated caution, and they pull the simmering argument over AI speed into the open. The dispute matters because the companies involved build the models that set the pace for the rest of the industry, and their public posture shapes how regulators and customers read the risk.

Anthropic has spent two years positioning itself as the safety-conscious counterweight to faster-moving competitors, publishing research on model behavior and urging deliberate deployment. Its leadership, including chief executive Dario Amodei, has repeatedly warned that capabilities are advancing quicker than the tools to control them. Zuckerberg's counter is blunt: he contends that framing restraint as principle can function as a commercial moat, letting a smaller player shape rules that bigger ones must follow. Meta, which releases open-weight models under its Llama line, has staked out the opposite bet, arguing that broad distribution spreads scrutiny and speeds up defensive research.

The two companies sit in very different places, which helps explain the friction. Meta can afford to give models away because its revenue comes from advertising across Facebook, Instagram and WhatsApp, not from selling model access. Anthropic sells API access and enterprise contracts, so its safety messaging doubles as a product story about reliability for corporate buyers. That structural difference turns a philosophical argument into a competitive one, and both firms know their public statements land with policymakers weighing new AI rules.

The timing sharpens the stakes. Governments in the United States, the European Union and Britain are still writing AI governance frameworks, and testimony from lab executives carries weight in those proceedings. If the industry cannot agree on whether to slow down, lawmakers may write rules that satisfy neither camp, or that favor the loudest voice. Investors are watching too, since the argument touches the core question of how quickly AI revenue can scale against the cost of building and running frontier systems.

What to watch next is whether Zuckerberg's criticism draws a formal response from Anthropic or stays a one-sided jab, and whether other labs pick sides. The bigger question is whether the slowdown debate hardens into two named camps, open-weights versus closed-and-cautious, with each courting regulators for legitimacy. For now, the disagreement is public, personal and unresolved, which is precisely why it will keep shaping how fast the next generation of models reaches users.