Dan Hendrycks calls for a pause on frontier AI, Max Hodak pushes back

What happened
At The New York Times' DealBook Summit in New York, Dan Hendrycks called for an indefinite halt to large-scale frontier AI training runs, arguing that this pause would allow time to improve safety and reliability. Max Hodak, also on the panel, countered that enforcing such a pause would require extensive surveillance measures. The discussion included other AI industry leaders and focused on regulation, global cooperation, and the challenges of aligning international policies. The panel also touched on issues of market concentration and the use of antitrust law as an alternative approach.
Key details
- Dan Hendrycks leads the Center for AI Safety and was a safety adviser to AI companies until recently.
- Hendrycks proposed exceptions for narrow models like AlphaFold and voice models, and said labs could still sell inference.
- Hodak argued that enforcing a pause would require an 'exhaustive surveillance state' on the internet.
- Lila Ibrahim of Google DeepMind, Mati Staniszewski of ElevenLabs, and Clem Delangue of Hugging Face also participated in the panel.
Why it matters
AI companies, policymakers, and researchers are directly affected by the debate over whether to pause advanced AI development. If a pause were implemented, companies would face operational limits on training large models, and governments would need to consider the costs and risks of surveillance to enforce such a ban. The discussion also highlights a concrete choice between regulatory intervention and using antitrust law to address concerns about safety and concentration of power in the AI sector.
