00:16A senior artificial intelligence researcher is issuing a chilling warning about the future
00:22of A.I. Evan Hubinger, alignment science lead at Anthropic, says he and his colleagues
00:29genuinely believe artificial intelligence could one day kill all humans. And, Hubinger
00:36puts his own estimate at more than 10% within the next decade. That is not a prediction
00:42that human extinction is inevitable. It is his personal assessment of what he considers
00:47a potentially catastrophic risk if increasingly powerful A.I. systems become difficult or
00:54impossible for humans to control. And, according to Hubinger, the industry does not yet have
01:00the solution. Hubinger says Anthropic is trying its best, but he also acknowledges a deeply
01:07unsettling problem. The company does not yet have a plan to solve what he calls alignment
01:12for superintelligence. And, he says Anthropic is not clearly on track to solve it. His biggest
01:19concern is what happens if A.I. begins improving itself, faster, smarter, and with less human
01:26oversight. The warning comes after another Anthropic researcher announced his resignation over A.I.
01:32safety concerns. Jacob Coxon, who said he previously worked on pre-training research at OpenAi and
01:39Anthropic, accused leading A.I. companies of racing toward self-improving superintelligence
01:46while taking potentially enormous risks. Coxon said people building these systems believe A.I.
01:52could kill us all by the end of the decade. And, he insisted that warning was not a marketing stunt.
01:59Coxon has also warned that the timeline could be far shorter than many people expect. He told the
02:05Wall Street Journal that some of the most aggressive scenarios could unfold rapidly, potentially putting
02:11the world out of control by the end of next year. In his online posts, Coxon compared his experiences
02:18at OpenAi and Anthropic. He argued that Anthropic employees understand the civilizational stakes better,
02:26but claimed the company remains locked in a race to develop increasingly powerful systems.
02:32Hubbinger says the immediate threat from current A.I. models is low, but that is not where his
02:37concern ends. His focus is on the possibility of recursive self-improvement — A.I. systems
02:44helping create newer and more capable versions of themselves. Hubbinger says this development may be
02:50happening faster than researchers expected. The warnings are not isolated. Several prominent A.I.
02:57researchers and executives have been pushing for greater caution as companies race to develop
03:02increasingly advanced systems. A July statement called Pacing the Frontier argued that industry,
03:09governments, and society may need the option to slow down long enough to develop stronger security
03:15measures and oversight. The statement was backed by prominent figures from Anthropic, OpenAi, and Meta.
03:21All right.
03:25All right.
03:51Subscribe to One India and never miss an update.
03:56Download the One India app now.
Comments