Ex-Anthropic: Slowing Down AI Would Be Ineffective Against Espionage

⚡

Key Takeaways

1Jacob Coxon, former Anthropic, deems slowing down AI ineffective, mentioning espionage and distillation

2Anthropic claims there is no evidence to support these suspicions and details its controls

3Laboratory leaders call for regulation of development, despite criticism regarding market impact

💡Why it matters — The debate over the pace of AI pits safety against international competition and economic issues.

Jacob Coxon, a former employee of Anthropic, believes that a slowdown in AI development would not produce the desired effect, citing industrial espionage and distillation attacks. Anthropic counters that there is no evidence to support these claims and emphasizes its internal controls. These statements come as leaders from major labs advocate for more regulated development.

Leaders Call for a Slowdown, Despite Criticism on Market Effects

Leaders from major AI labs, including Dario Amodei, Sam Altman, and Elon Musk, have recently discussed the need to slow down AI development and add safeguards. Sam Altman indicated in September on X that progress at OpenAI has been rapid and will continue to be, but it will be slightly slowed down to enhance safety. Meanwhile, many critics believe that such a slowdown could create a buzz around the anticipated IPOs of Anthropic and OpenAI, while disadvantaging smaller competitors in the sector.

Coxon Argues That a Slowdown Would Be Ineffective Against Espionage

In an interview published Tuesday with Patrick Bet-David, Jacob Coxon stated that a slowdown in AI development in the United States would be ineffective, citing the possible presence of Chinese spies in American labs. He expressed 90% certainty that such spies exist at OpenAI and Anthropic, capable of seizing advancements and allowing China to catch up. According to him, whether the U.S. continues or slows its work, China would eventually gain access to these technologies, either through leaks or independent development. When asked about the identity of these spies, Coxon clarified that he did not want to initiate a witch hunt and did not name anyone.

Anthropic Disputes the Allegations and Details Its Controls

According to a source close to Anthropic, there is no evidence to support Jacob Coxon's accusations regarding potential Chinese spies. Anthropic emphasizes that it considers the risks associated with foreign interference to be significant and assures that it implements a strict employee verification process, rigorous access restrictions, and internal risk monitoring.

Distillation: A Pathway to Progress Pointed Towards China

Jacob Coxon also mentioned the use of distillation attacks as a factor in the advancement of AI capabilities in China. These attacks involve using advanced models to train and improve less capable models. This summer, Anthropic accused Chinese labs, including Alibaba, Moonshot, DeepSeek, and Zhipu, of launching millions of such attacks.

Background and Internal Culture Discussed by Coxon

Jacob Coxon left Anthropic last month after accusing Anthropic and OpenAI of "playing with lives." He also compared Anthropic's staff to worker ants ready to sacrifice themselves for the mission of achieving superintelligence.