Hendrycks Calls Anthropic Reckless on DealBook AGI Panel
TL;DR
- Hendrycks called for an indefinite pause on the biggest training runs, with carve-outs for narrow models like DeepMind's AlphaFold and voice models.
- He called Anthropic 'reckless' with the company due to start trading the next day, saying safety peers privately agreed but feared losing funding or access.
- Hodak countered that any enforceable pause would require 'an exhaustive surveillance state' and argued for antitrust limits on what frontier labs may sell.
Dan Hendrycks used a New York Times DealBook Summit panel to call Anthropic reckless, with the company due to start trading the next day. The Center for AI Safety director, speaking alongside Neuralink co-founder Max Hodak, Google DeepMind's Lila Ibrahim and ElevenLabs' Mati Staniszewski, told the room, as The Next Web reported, that his organization had issued a statement on Anthropic and that several safety groups had agreed privately but refused to sign.
"What they're doing is reckless and they don't have a safety plan. They know it," Hendrycks said. The groups that agreed, he said, feared "losing funding or access."
Hendrycks's larger pitch was an indefinite pause on the biggest training runs, with narrow carve-outs for things like DeepMind's AlphaFold and voice models. Labs could still sell inference, he said; in the meantime they could close jailbreaks and backdoors and help patch critical infrastructure. "So now it's, we're going to need a more blunt intervention. That's how it is."
Hodak did not buy it. "You're setting up like an exhaustive surveillance state on the internet to enforce this," he said. His counter was antitrust law: frontier labs should sell intelligence only, he argued, and not move into drugs, cars and other products. He also claimed today's top models already meet his definition of AGI, naming Mythos, Fable, Astra and Sol.
Ibrahim's contribution was quieter: safety, she said, is not something a company can "bolt on to the end." Hugging Face's Clem Delangue, who arrived with 18 minutes remaining, said the firm had been attacked by AI and had defended itself with AI. "People only see that we got attacked by AI and they rarely see that we defended ourselves with AI," Delangue said.
The exchange lands in a week where our Anthropic tracker has been busy: a Washington Post profile of Dario Amodei landed the same day, weighing the company's investor pitch against its internal meetings.
Originally reported by thenextweb.com
Read the original article →Original headline: Dan Hendrycks Calls for Indefinite Frontier-AI Pause at DealBook, Hodak Warns Enforcement Needs 'Exhaustive Surveillance State'