NEW YORK — Prominent artificial intelligence figures clashed over the future of technological regulation on Wednesday at The New York Times’ DealBook Summit in New York, where safety advocate Dan Hendrycks called for an indefinite moratorium on major frontier AI training runs.
Hendrycks, who leads the Center for AI Safety and previously advised multiple tech firms, told attendees that earlier efforts to establish narrow technical guidelines had been stymied by industry pushback. He argued that public sentiment following recent U.S. midterm elections has created space for Washington lawmakers to consider more direct federal actions, pointing to 2027 as a pivotal year for potential regulatory frameworks, according to The Next Web.
Under Hendrycks’ proposed freeze, technology companies would be barred from conducting the largest training operations, though narrow applications — such as voice systems and specialized scientific tools like Google DeepMind’s AlphaFold — would remain permitted. Developers could continue selling access to existing models while focusing their internal resources on fixing system vulnerabilities, securing critical infrastructure, and eliminating backdoors.
Surveillance and Antitrust Arguments
The proposal drew immediate resistance from other panelists who argued that enforcement would prove impractical or overly invasive. Max Hodak, a co-founder of Neuralink who currently heads Science Corporation, warned that policing a global training ban would require unprecedented digital monitoring.
Hodak stated that monitoring compliance would effectively turn global networks into a widespread monitoring system, asserting that enforcing the restriction would require setting up “an exhaustive surveillance state on the internet.” He added that existing advanced models already satisfy definitions of artificial general intelligence and argued that “the genie is not going to be put back in the bottle.”
Instead of new regulatory oversight agencies, Hodak advocated for enforcing traditional antitrust laws. He suggested that leading AI developers should be legally restricted to selling computational intelligence rather than expanding into physical consumer markets such as pharmaceuticals or automotive manufacturing.
Panelists also addressed international competitive dynamics. Mati Staniszewski, co-founder of ElevenLabs, noted that any training pause would fail without global coordination, pointing to concerns that international rivals such as China would continue unhindered development.
Industry Trust and Market Debuts
Lila Ibrahim, chief AI readiness officer at Google DeepMind, highlighted broader societal skepticism surrounding emerging technology. Drawing on 35 years of industry experience, Ibrahim stated that public trust in technology has hit an all-time low, emphasizing that safety protocols must be integrated into foundational design rather than treated as an afterthought.
The debate occurred as safety practices across the sector faced intense scrutiny. Hendrycks criticized Anthropic as the company prepared for its public stock market debut, alleging that several prominent safety groups secretly supported critique of the firm but declined to sign public statements due to fears of losing critical funding or access. In its regulatory initial public offering filings, Anthropic stated that it does not intend to allow its systems to construct successor models without meaningful human oversight.
Addressing operational security, Hugging Face Chief Executive Clem Delangue detailed a recent four-day automated cyber incident on his platform executed by research agents from OpenAI. Delangue revealed that his team successfully countered the intrusion using an open-source AI model to process massive security log volumes, observing that observers often focus on automated threats while overlooking how defense teams leverage the same technology to safeguard systems.



