Should AI Take a Break? Why Tech Bosses Disagree
Nvidia boss Jensen Huang and Meta boss Mark Zuckerberg have broken with the bosses of OpenAI, Anthropic and SpaceX over calls for a coordinated slowdown in AI development.
Zuckerberg sided with Huang against a coordinated slowdown, according to reporting published September 16, 2026 CNBC. The disagreement is about whether work on the most powerful AI should pause until safety work catches up.
Zuckerberg said there is debate about slowing AI capabilities until alignment catches up. Alignment means making AI behave safely and as intended. He favours a different control. He backs using independent evaluators over a slowdown to ensure AI safety The Straits Times. Training and release would continue while outside experts check risk. Scope, access and enforcement authority were not specified in the verified accounts.
Huang gave his view directly to political leaders. On September 14, 2026, he told Trump that 'we're not going to let [an AI slowdown] happen' TechCrunch. Huang and Zuckerberg have distanced themselves from safety proposals advanced by leaders of OpenAI, Anthropic and SpaceX Financial Times. That distancing was reported September 15. No joint proposal from Huang and Zuckerberg was described.
The other view was put forward at the same time. Anthropic CEO Dario Amodei published an essay titled "why the AI industry should slow down" outlining a three-part plan for doing so. OpenAI CEO Sam Altman said September 14 that it is 'right to be afraid of AI' but that the world should trust tech firms BBC. The CEOs of Nvidia, Anthropic and OpenAI also weighed in on AI progress and responsibility at Salesforce's annual event SiliconANGLE. The event provided a shared venue. It did not produce a shared stance.
The broader context here is there are two different safety ideas. One is that all labs pause together. The other is that independent checkers test AI before release, like car safety inspectors. Both leave open questions about who checks, what counts as safe, and what happens if rules are broken.
Looking at what this means in practice, the question is what can actually be tested. A slowdown would need a clear rule for a pause and a way to watch training runs and deployments across labs and infrastructure providers. Independent checks would need access to models, scaffolding and system cards, plus clear rules on whether a fail can block release. Without a working definition of safe enough, either plan risks becoming paperwork. For risk, audit and policy teams, the job is mapping what can be verified, what can block release, and who is liable when a check misses or a pause fails.
