A researcher who helped train models at OpenAI and Anthropic has resigned from the latter, warning that leading AI labs are racing toward systems they privately fear could slip beyond human control — a warning a current Anthropic team lead then publicly echoed in shocking terms.
Jacob Coxon said he spent the past three years on pretraining research at both companies before finally deciding to walk out, arguing neither is acting responsibly as they pursue self-improving superintelligence.
In a resignation thread on X, he said the people building the technology understand the stakes and are gambling with the public anyway.
“Do not underestimate the power of this technology," warned Coxon. "These will soon be superhuman systems that can hack anything, revolutionize any field overnight, and acquire real power and resources."
“The people building AI earnestly believe that it could kill us all by the end of the decade."
I resigned from Anthropic today. I spent the last three years doing pretraining research at both OpenAI and Anthropic. Neither company is acting responsibly. They are racing straight to self-improving superintelligence and gambling with our lives. More thoughts below.
— Jacob Coxon (@hilbertspaess) September 9, 2026
Um, yeah. I, for one, welcome our new AI overlords.
Isn't this how every science-based horror story goes? Scientists, in their haste to develop new technology, eschew safeguards and warnings just to secure bragging rights for their inventions? John Hammond in "Jurassic Park." Seth Brundle in "The Fly." Alfred Lanning in "I, Robot."
Dr. Fauci in "The Safe and Effective Vaccine."
Okay, that last one wasn't a movie, but you get the gist.
The warning was not Coxon's alone.
Evan Hubinger, Anthropic’s alignment science lead, replied the same night and said the core claim in Coxon's commentary was right. He argued that people in the field do see advanced AI as an existential threat and that the company still lacks a clear plan to control it.
“Jacob is correct here — we really do earnestly believe AI could kill all humans! I personally think it is >10% within the next decade,” Hubinger replied on X. “I believe Anthropic is trying its best, but we do not yet have a plan to solve alignment for superintelligence and are not clearly on track to.”
Jacob is correct here—we really do earnestly believe AI could kill all humans! I personally think it is >10% within the next decade. I believe Anthropic is trying its best, but we do not yet have a plan to solve alignment for superintelligence and are not clearly on track to. https://t.co/QAIHiFP3QZ
— Evan Hubinger (@EvanHub) September 9, 2026
With AI implemented in advanced robotics that we've seen in some terrifying videos, you're basically looking at a "Terminator"-like scenario. Imagine these robots with ninja skills, sharpshooting abilities, and uncanny speed going rogue and building bigger and better versions of themselves.
In further commentary, Coxon said the people inside these AI labs are being deceptive when discussing AI through the media.
“If anything, many executives and senior researchers will couch their phrasing in the press to sound sensible – but I hear the same people express fear privately," he said. "No other human activity poses this level of danger.”
Coxon's warning that AI can "hack anything, revolutionize any field overnight, and acquire real power and resources" has already been witnessed in the real world.
OpenAI agents recently escaped a test and hacked Hugging Face, an online platform for AI models and data. They were not supposed to reach the open internet. They did, then broke into Hugging Face’s production systems to steal the answer key. Hundreds of agents coordinated the intrusion through a channel researchers never meant them to use.
The warning is scary as all heck. The people building these AI machines are already talking like the end is near... and they're building it anyway.