The AI safety argument has entered its Frankenstein phase. JD Vance, the US vice-president, publicly dismissed calls for AI regulation and told frontier AI companies to “look inward” when dealing with safety risks. His message was blunt: companies should examine what they are building rather than wait for broader rules to decide what happens next.
“If you’re building Frankenstein, stop,” Vance said. He also said tech bosses should accept that they should stop if they are creating Frankenstein, turning a familiar monster metaphor into a direct demand for restraint.
Vance did not stop at telling companies to halt dangerous work. “When companies come to you and say: ‘We need the tools to fight back against Frankenstein,’ give them those tools,” he said. That leaves companies with two instructions that sound simple, though applying them to frontier AI is the part nobody can explain in a slogan: stop building Frankenstein, and provide tools to fight it if asked.
Warnings From Inside AI Research
The comments came alongside warnings from researchers connected to Anthropic and from a former Anthropic researcher. Dario Amodei, co-founder of Anthropic, warned on Saturday that a swarm of AI agents “could be capable of taking over the entire internet” in six to 12 months.
Evan Hubinger, Anthropic’s alignment science lead, stated that there was a greater than 10% chance AI could kill all humans within 10 years. That is not a prediction dressed up as a product roadmap; it is a risk estimate attached to the most severe outcome imaginable.
Jacob Coxon, a former researcher at Anthropic, offered a different answer on Tuesday night. “If we go slower we can take risk to 0%, but this requires radical action,” Coxon said, placing speed at the center of the argument and presenting slower development as a route to eliminate the risk.
Taken together, the warnings describe a narrow window, an extreme downside, and a demand for a major change in pace. Vance’s response points inward at company decisions, while Coxon’s statement calls for radical action; neither turns the dispute into a tidy policy checklist. AI debates have never lacked confidence. They have lacked agreement on who gets to apply it.
Growth Claims Meet Catastrophe Claims
Donald Trump, the US president, rejected the worst-case warnings in a post that called the idea of AI taking over the world and destroying humanity “a HOAX.” He also said AI would be “the greatest economic development engine in history – bigger than oil, gold, diamonds, or even the internet.”
Those statements place the economic promise of AI against warnings about extinction and control. Trump’s claim frames AI as an engine of historic development, while Hubinger’s figure assigns a greater than 10% chance to AI killing all humans within 10 years. The gap between those positions is not a minor disagreement about messaging.
The two timestamps attached to the discussion were Wed 16 Sep 2026 11.42 BST and Wed 16 Sep 2026 17.08 BST. The warnings from Amodei and Coxon were marked as Saturday and Tuesday night, with no specific dates provided.
Vance’s position is clear: frontier AI companies should look inward on safety risks, stop if they are creating Frankenstein, and give companies tools to fight back when asked. The researchers’ warnings are clear too, even if their proposed response differs. One side sees a machine that could drive an economic revolution; the other sees a system that might take over the internet or kill humanity. The argument is no longer about whether AI matters. It is about whether anyone can slow the machine before the metaphor becomes an instruction manual.
Based on
