An Anthropic researcher who resigned over fears that artificial intelligence could destroy humanity says the problem is solvable — if governments step in before the industry races itself into extinction.
Jacob Coxon, 27, who spent the past three years doing pretraining research at OpenAI and Anthropic, told Fox News on Wednesday that companies are not acting responsibly. His resignation, posted Tuesday on X, warned that the race toward self-improving superintelligence is “gambling with our lives.”
Coxon spoke with Fox News chief political anchor Bret Baier on “Special Report.” He said regulation is key.
Coxon’s resignation warning
Coxon said the immediate threat from current AI models is “minimal.” The danger comes later, when AIs start making themselves smarter — a process called recursive self-improvement.
“But recursive self-improvement — this AIs making themselves smarter — could happen as soon as next year,” Coxon said. “Some people even say six months. So this attitude is shared across the industry, where people essentially are aware that the work they are currently doing, if we don’t slow down, could lead to AI takeover by the end of the decade.”
He believes there is “definitely” something the country can do to change the dynamic.
“I think the problem is not unsolvable, it’s not like we think this is some impossible thing — it’s just, we’re racing so fast, we haven’t got time to solve it correctly, we’re not being careful, and everyone wants to be careful,” Coxon said. “I think it’s just — people are begging for regulation. I think it’s just a matter of — international — somebody stepping in and being able to ensure that we don’t race ourselves into extinction. The problem is eminently solvable.”
Hubinger’s agreement
Evan Hubinger, alignment science lead at Anthropic, publicly backed Coxon’s post.
“Jacob is correct here — we really do earnestly believe AI could kill all humans!” Hubinger wrote on X. “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.”
Hubinger added: “To be clear, as we say in our latest Risk Report, I think the risk from present models is low. What I am worried about is superintelligence arising from recursive self-improvement, as we have said is happening faster than we thought.”
The ‘superweapon’ arms race
Coxon called AI “possibly the most dangerous technology that humanity has ever created” and said it is being used as a “superweapon” in an international arms race. He called for global oversight.
“It’s basically out of science fiction. I think we have no other choice but to cooperate internationally because an arms race would be disastrous in a way that no other human activity has been in the past.”
He said technology leaders have publicly asked for a slow-down agreement, but people are not taking them seriously.
“It’s not just a marketing stunt,” he said. “These people are deadly serious that they think their industry needs to be regulated as soon as possible.”
A solvable future
Coxon said we are “on the doorstep of tremendous discoveries,” but the industry is moving too quickly. He emphasized moderation.
Baier asked Coxon if the issue could be figured out.
“I think definitely. I think if we actually try instead of submitting to the inevitability of a doomed race or a hubristic gamble, then we can solve it,” he said. “The problem is not unsolvable.”
Fox News Digital reached out to Anthropic for comment.
The numbers
- Coxon, 27, spent three years doing pretraining research at OpenAI and Anthropic
- He posted his resignation Tuesday on X
- Hubinger estimates a >10% chance AI kills all humans within the next decade
- Coxon says recursive self-improvement could arrive as soon as next year
What Coxon wants
- International regulation of AI development
- A slowdown agreement among technology leaders
- Global oversight of AI, which he described as being used as a “superweapon” in an arms race
Source: foxnews.com
Get the Notebook.
The day's best stories and every fresh verdict, in plain English, in your inbox by seven. One email a day, no more.