Dan Hendrycks, director of the Center for AI Safety, urged an indefinite halt on major AI training runs to improve safety and allow regulatory progress, while Max Hodak countered that such a pause would demand invasive surveillance and stifle innovation.
- Hendrycks calls for an indefinite pause on the largest frontier AI training runs.
- Hodak argues a pause would require a draconian surveillance state to enforce.
- Panelists agree regulation will be a key focus in AI's evolving global landscape.
What happened
At The New York Times’ DealBook Summit, Dan Hendrycks proposed an indefinite pause on significant frontier AI model training to address critical safety and reliability issues. Hendrycks suggested exemptions for niche models like AlphaFold and voice recognition AI, with labs allowed to continue inference services while enhancing security.
Max Hodak, co-founder of Neuralink and CEO of Science Corporation, challenged the feasibility of such a pause, warning it would require a pervasive surveillance mechanism reminiscent of a police state. Hodak argued that top AI models already reach artificial general intelligence (AGI) standards and that safety comes from broad technology distribution rather than containment.
Why it matters
The discussion underscores the growing tension between AI innovation and governance, with the 2027 political environment opening avenues for more direct regulatory actions. Hendrycks sees political appetite for blunt interventions due to past industry resistance to narrow rules and rising public concern.
However, the technological and geopolitical complexities involved — such as global alignment, enforcement difficulties, and competitive pressures from nations like China — present significant barriers. Hodak and others emphasize antitrust measures and widespread diffusion of AI capabilities as alternative paths to ensuring safety and preventing concentration of power.
What to watch next
Expect heightened regulatory activity on AI in 2027, with policy debates focused on whether to impose training moratoriums or pursue alternative controls like governance frameworks and market competition safeguards. How governments balance innovation incentives and public safety will be critical.
Global coordination remains uncertain but essential; without it, unilateral pauses risk ineffectiveness. Industry players like Google DeepMind and ElevenLabs, and stakeholders such as Hugging Face, will play crucial roles in shaping norms. Meanwhile, companies going public, such as Anthropic, will face scrutiny over safety commitments amid rapid development.