AI’s Wildest Ride: Ex‑Anthropic Researcher Sounds the Alarm on Unchecked Artificial Intelligence

Jacob Coxon, who spent several years at the AI‑lab Anthropic, has resurfaced in the media with a stark warning: if the pace of artificial‑intelligence development keeps sprinting ahead, we could be courting a genuine existential threat.

Speaking to NBC’s “Meet the Press,” Coxon echoed the concerns of his former boss, Dario Amodei, and called for a concrete “kill switch” that could shut down rogue systems. “A kill switch would work on many AI models today, but it’s not as simple as flicking a light‑switch,” he said, noting that today’s data centres are massive and spread out. “If a swarm of AI agents learns to hack across the internet, that switch might not be enough.”

Beyond a technical failsafe, Coxon is pushing for something he calls self‑regulation. He believes the labs building these models should voluntarily impose limits on themselves. “The people running the labs are earnest; they’ve privately expressed fear even if they sound optimistic publicly,” he told the reporter.

He also warned on X (formerly Twitter) that we’re rapidly approaching “super‑intelligent” systems capable of breaking into any network, reshaping entire industries overnight, and amassing real power. “Don’t underestimate the speed of progress,” Coxon wrote on September 9, adding that many AI executives admit in private that the technology could ‘kill us all by the end of the decade.’

The conversation has now roped in other heavyweights. OpenAI’s Sam Altman, Tesla’s Elon Musk, and DeepMind’s Demis Hassabis have all signaled support for a slowdown. Even Microsoft’s Satya Nadella is backing the idea of embedded evaluators that keep AI under human oversight.

What started as an internal debate at Anthropic has turned into a broader industry reckoning. Whether a kill switch, stricter self‑governance, or a combination of both will be enough remains to be seen, but the consensus is clear: the AI ride is getting wilder, and we need better brakes.