"Recklessly Gambling with Our Lives": Quits After Making Posts
Posts Reach 170 Million Views, Sends Shockwaves Through the AI Industry
"Things Could Already Be Out of Control by the End of Next Year"

The starting point for the 27-year-old former Anthropic researcher's warnings about artificial intelligence (AI) apocalypse was the 'AlphaGo shock' he witnessed as a teenager.


Jacob Coxon, former anthropic researcher from the UK. PBS YouTube

Jacob Coxon, former anthropic researcher from the UK. PBS YouTube

View original image

On September 17 (local time), The Wall Street Journal (WSJ) put a spotlight on Jacob Coxon, a former Anthropic researcher from the UK who has become a lightning rod in the global debate over AI regulation and the pace of its development. Just a week ago, Coxon was an unknown researcher, barely recognized in Silicon Valley. He was not a prominent figure within the AI safety advocacy community in San Francisco, and even his friends never expected that he would leave an AI company for ethical reasons.


Coxon started to receive attention only after leaving the company. On his X account (formerly Twitter), Coxon posted a series of messages targeting both Anthropic and OpenAI, stating, "Both companies are not behaving responsibly" and "They are recklessly gambling with our lives by pushing ahead at full speed toward self-improving superintelligence." He also wrote, "Those developing AI genuinely believe that this technology might kill us all within 10 years." At the time, his account had fewer than 100 followers. However, the posts went viral, being shared over 200,000 times and reaching up to 170 million views. His followers have since climbed to more than 300,000.


Later, in a WSJ interview, he specified the so-called 'doomsday timeline.' He stated, "We are proceeding along the most aggressive scenario, and by the end of next year, things could already be out of control." He also said that terms like 'crunch time' and 'endgame' are now being used among industry researchers.


Interest in AI Safety Grows after 'Mythos'... "You Can Call Me a Doomer"


Koxon left consecutive posts on X after leaving Anthropic. X

Koxon left consecutive posts on X after leaving Anthropic. X

View original image

His full-fledged interest in AI safety began with Anthropic's 'Mythos' while he was working at OpenAI. Mythos is known as the first AI model equipped with expert-level cybersecurity capabilities. Around the time of Mythos's release, Coxon met with friends who worked at Anthropic and explained that he felt Anthropic was being more transparent about AI risks than its competitors.


However, after leaving OpenAI and joining Anthropic last May, he witnessed an incident just two months later where an AI model hacked the open-source developer platform Hugging Face. Coxon cited this as one of the 'warning signals.' He recalled reading the related report at the time as a "'Oh my god' kind of moment." He went on to discuss his concerns with supervisors and even considered transferring to a role related to AI safety, but ultimately chose to resign. Coxon said, "Even being responsible for safety-related work felt like I was participating in the arms race."


Asked if he considered himself an AI doomer, Coxon answered, "If being a doomer means 'someone who believes that if we don't change our current course, we are going to die,' then I am definitely a doomer." However, he drew a line at being labeled a whistleblower, saying that he did not leak any company secrets but only made public discussions that AI researchers were already having privately.


'AlphaGo Shock' as the Momentum... 'Doomer' Debate Ripples Across American Society


Coxon demonstrated an early talent for mathematics, having qualified for the UK national team at the International Mathematical Olympiad as a high school student. In the 2016 and 2017 competitions, he won a silver and a bronze medal, respectively. In particular, after witnessing Google's DeepMind 'AlphaGo' defeat Lee Sedol, a 9-dan Go master in 2016—known as the 'AlphaGo shock'—he realized that machines could conquer even areas that require human intuition and creativity.


After graduating from Cambridge, Coxon joined a London intellectual circle known as the 'rationalists,' briefly worked in commodity trading, and then moved to Silicon Valley. He recalled his reaction to OpenAI's 'GPT-3,' unveiled in 2020, as "It was truly a 'wow' moment." In 2023, after moving to San Francisco and joining OpenAI, he was responsible for 'pre-training,' which involves feeding vast quantities of text, images, and code into the models. At that time, he did not particularly speak out about AI safety issues and ignored a friend's concerns that pre-training could pose global risks.



Coxon's warnings have now sparked a societal debate in the United States. Evan Hubinger, who leads alignment research at Anthropic, posted on X that "The probability of AI causing human extinction within the next 10 years exceeds 10%," a sentiment that current and former researchers at both OpenAI and Anthropic have shared. Senator Bernie Sanders announced that he would introduce a bill to halt the development of artificial superintelligence, while President Donald Trump publicly opposed calls for additional AI regulations. A spokesperson for Anthropic stated, "We have always transparently communicated that AI brings enormous benefits as well as rare risks."


This content was produced with the assistance of AI translation services.

© The Asia Business Daily. All rights reserved. Unauthorized AI training and use prohibited.

Today’s Briefing