Meta (META.US) Zuckerberg Responds to the AI "Slowdown" Debate: Continue Advancing Model Capabilities, Strengthen Safety Protections

Meta Platforms (META.US) CEO Mark Zuckerberg and the company's AI chief Alexander Wang have weighed in on the debate over the pace of AI development, arguing that AI labs should continue advancing model capabilities while building stronger safeguards around increasingly powerful models. The two made their remarks on X, amid a broad industry debate over whether capability development should slow down until more progress is made on AI safety ensuring AI systems reliably follow human intent and values. Meta shares rose for a third straight session through Tuesday, gaining 4.2% cumulatively, and edged higher in after-hours trading. On Stocktwits, retail sentiment on the stock was "extremely bearish" as of the latest check. **Zuckerberg: Labs Have Incentive to Build Safer AI** Zuckerberg said AI companies already have strong commercial and legal incentives to make models safe and aligned with users. "People won't want to use agents that aren't aligned with them and don't do what they ask," Zuckerberg wrote. "Any lab that doesn't focus on alignment will fall behind." He argued that alignment itself is becoming a key competitive capability, rather than merely a constraint on AI development. "My view is that trust and alignment are rapidly becoming the most important capabilities that distinguish agents and models," he said. Zuckerberg also noted that Meta once decided to delay the release of its Muse model by several months to address safety and security issues. **Meta Backs External Oversight** Both Zuckerberg and Alexander Wang called for greater involvement of independent evaluators and external oversight as AI models grow more powerful. "Bringing in independent evaluators and advisors is industry best practice," Zuckerberg said, arguing that a larger, more diverse ecosystem of evaluators will improve AI safety work. Alexander Wang likewise said AI labs should maintain "strong governance frameworks" throughout the training and deployment pipeline, including external evaluators and independent oversight of safety standards for model releases. He added: "Every lab must operate under the institutional protections of democratic nations... This means labs will face significant legal liability if their models cause harm." **Meta Defends Compute Allocation Strategy** A key part of the two executives' argument is Meta's decision to devote most of its compute to serving users rather than accelerating recursive self-improvement (RSI). "Putting the vast majority of compute into serving humans rather than racing toward recursive self-improvement is one of the best ways to ensure we develop this technology safely," Zuckerberg said. The concern is that RSI could allow AI systems to rapidly improve themselves, potentially outpacing human oversight and increasing the risk of losing control of powerful models. Alexander Wang similarly described racing toward RSI as "one of the most dangerous paths that could lead to loss of control of powerful models." **AI Pace Debate Polarizes Tech Industry** The debate over the pace of AI development broadly splits into camps: Anthropic CEO Dario Amodei has called for slowing frontier AI development so safety mechanisms can keep up; OpenAI CEO Sam Altman, Microsoft Corporation CEO Satya Nadella and SpaceX's Elon Musk support a more cautious, safety-focused approach without halting development. On the other side, U.S. President Donald Trump opposes slowing the American AI race, citing concerns about China pulling ahead; former Trump AI advisor David Sacks has said companies should manage risks themselves rather than seek government restrictions, and has questioned Anthropic's safety warnings. Investor Michael Burry has questioned whether calls to slow development are partly self-interested, as they could shield incumbent AI companies from competitive threats.