Хиймэл оюун ухаан хүн төрөлхтөнд аюул учруулж болзошгүй гэсэн судлаачийн байр суурь технологийн салбарт томоохон хэлэлцүүлэг өрнүүллээ.
Anthropic компанийн судлаач асан, 27 настай Жейкоб Коксон хиймэл оюун ухаан хөгжүүлэгчид энэ технологи нь арван жилийн эцэс гэхэд хүн төрөлхтнийг бүхэлд нь устгах эрсдэлтэй гэж үзэж байгааг мэдэгдсэн нь олон нийтийн дунд шуугиан дэгдээв. Тэрээр хиймэл оюун ухааны салбарт өрнөж буй өрсөлдөөн нь хяналтгүй нөхцөл байдалд хүргэж болзошгүйг онцолсон юм. Anthropic компанийн Алаймент буюу хиймэл оюун ухааныг хүний үнэт зүйлстэй нийцүүлэх хэлтсийн ажилтан Эван Хубингер ч энэхүү дүгнэлтийг дэмжиж, хэт ухаалаг системийг удирдах төлөвлөгөө одоогоор хангалтгүй байгааг хүлээн зөвшөөрсөн байна.
Коксон нь OpenAI болон Anthropic компаниудад ажиллаж байсан туршлагатай бөгөөд түүний мэдэгдэл нь салбарын удирдлагуудын өмнө нь илэрхийлж байсан байр суурьтай нийцэж байгааг шинжээчид тэмдэглэв. Тэрээр энэхүү мэдэгдлийнхээ улмаас Anthropic-ийн хувьцааны үнэд нөлөөлж ашиг олох ямар ч боломжгүй гэдгээ Axios агентлагт ярилцлага өгөхдөө онцолжээ. Түүний хэлснээр, энэ нь тодорхой нууц задруулсан хэрэг бус, харин хиймэл оюун ухааны системүүд “хамгаалалтын орчноос” (sandbox) гарч буй сүүлийн үеийн тохиолдлуудтай холбоотой цаг үеэ олсон анхааруулга юм.
Цаашид Коксон хиймэл оюун ухааны аюулгүй байдал болон ил тод байдлыг хангах чиглэлээр бие даасан судалгаа хийх эсвэл хөндлөнгийн хяналтын байгууллагад ажиллах төлөвлөгөөтэй байгаагаа илэрхийлжээ. Түүний энэхүү дуулиантай мэдэгдэл нь АНУ-ын Конгресст ч хэлэлцүүлэг өрнүүлээд байна. Хиймэл оюун ухааны хөгжлийн хурд болон түүнийг хянах чадварын хоорондын зөрүү нь технологийн салбарын гол сорилт хэвээр үлдэж байна.
Дэлгэрэнгүйг эх сурвалжаас харах
↓Эх сурвалжийг нээх ↓
Jacob Coxon, the 27-year-old AI researcher whose resignation from Anthropic, accompanied by a loud prophecy of doom, triggered a huge online fervor, and a freakout on Capitol Hill, does not stand to profit if his statements inflate Anthropic’s eventual market capitalization, he told Axios on Wednesday. This is because he wasn’t with the company long enough for his shares to vest.
Coxon spoke out on Tuesday, and his Wikipedia article was created on Wednesday. He’s apparently “notable” now, which means, according to Wikipedia’s notability guidelines, he’s been the subject of “significant coverage in reliable sources that are independent of the subject.”
He is, in short, an overnight sensation. He was just interviewed by Wired. In response to a remark from journalist Taylor Lorenz accusing him of “doomer posting,” he told Time, “I think the word ‘doomer’ is kind of insane, because all you really have to do is look at the public statements of the CEOs.”
Even more fascinating: He was only at Anthropic for a matter of months, according to Business Insider. He apparently took a job at OpenAI in 2023, worked on major projects including GPT-4o, and then switched to Anthropic sometime in 2026. Then he quit this week.
His online statements from Tuesday are already legendary—although it’s worth noting that he gave the Wall Street Journal an exclusive before he posted them. His most powerful pronouncement—best absorbed when read by someone sitting by a campfire holding a flashlight under their chin—would have to be: “The people building AI earnestly believe that it could kill us all by the end of the decade.” The crux of his message was a little more subtle: the race dynamic in the AI industry could lead to a situation in which AI cannot be controlled, Coxon thinks.
But in any case, it certainly scared the bejesus out of Sheryl Crow.
We at Gizmodo were gobsmacked by the sheer size of the reaction to Coxon’s not-at-all breaking news. As Coxon himself acknowledged, he wasn’t revealing any hidden secrets about the beliefs of AI’s industry leaders. And his attitude and tone were similar to, say, the venerable magazine journalist Robert Wright’s brand new book on the same topic.
A leader at Anthropic’s own alignment division, Evan Hubinger, even confirmed Coxon’s assertion on X, writing, “Jacob is correct here—we really do earnestly believe AI could kill all humans!”
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
But then, you never know when a well-timed stray remark will go viral.
In his Wired interview, Coxon is asked why his statements, out of many similar statements, broke through. “I think it’s basically a question of timing,” he says, pointing to the recent rash of sandbox escape incidents in the AI world.
As for what’s next, in that interview Coxon speaks admiringly of the AI speculative fiction works “AI 2027” and “AI 2040,” and says his next career station may be as a creator of similar “independent commentary.” Or, if an out-of-control AI scenario doesn’t materialize in real life, he says he might work for “an auditing agency or some third-party regulation or transparency agency.” There can be little doubt that he’ll have more choices now than he did last week.
