Urgent.News

What's breaking now, across hundreds of outlets.

AI

Jacob Coxon says this is why pacing the frontier might not work

Coxon said in an interview that he was "almost certain" Chinese spies had infiltrated US AI labs, making it difficult to keep AI advances from China.

AI researcher Jacob Coxon argued that slowing AI advancement might not be a reliable solution to safeguard against potential threats. Jacob, a former Anthropic employee, asserted his 90% certainty regarding Chinese spies infiltrating US AI labs, capable of stealing progress and gaining a competitive edge. Jacob's statement came during an interview on businessman Patrick Bet-David's podcast.

He contended that either continuing AI development would expose the labs to Chinese spies, or halting it would enable China to independently develop the technology. Despite the lack of evidence supporting this claim, Anthropic maintains stringent security measures, including personnel vetting, access controls, and insider risk monitoring.

Coxon also pointed out the risk of distillation attacks, where advanced AI models enhance less sophisticated ones, a tactic employed by various Chinese labs, including Alibaba, Moonshot, DeepSeek, and Zhipu. He likened Anthropic employees to diligent worker ants, devoted to the success of the mission, striving towards superintelligence.

The interview unfolded as executives from leading AI companies, such as Anthropic's Dario Amodei, OpenAI's Sam Altman, and Elon Musk of xAI, discussed the necessity to moderate AI progress and incorporate safety measures. However, critics argue that slowing AI development could generate hype around the anticipated IPOs of Anthropic and OpenAI, potentially disadvantaging smaller AI rivals.

Rewritten by urgent.news; may contain errors — check the original.

Source: businessinsider.com

Read the original at businessinsider.com →

Report a problem with this story

More in AI

I let an AI agent drive my remote desktop. It can't touch my passwords.

It's 11 PM. Three servers need the same patch, and I'm alt-tabbing between four remote windows. Two are RDP, two are VNC.

  • AI agent can view, click, type, and navigate remote desktop.
  • Agent never accesses user passwords or credentials.
  • Agent control limited by time, can be reclaimed instantly.

Where AI excels—and where it quietly fails you

Artificial Intelligence is a powerhouse for productivity and complex tasks, but relying on it for emotional advice or personal coaching reveals its greatest limitation: pleasing you instead of truly helping you.

A theoretical computer scientist, citing sources, says AI labs have quietly started probing whether their models can break important cryptographic protocols (Scott Aaronson/Shtetl-Optimized)

friend-of-the-blog Omer Reingold (shared with permission) — Last night my 9-year-old son was taunting my wife …

  • AI labs probing cryptographic protocols, as per Scott Aaronson
  • 372 groundbreaking results from OpenAI, including UGC proof
  • Complexity of understanding AI-generated proofs in mathematical community

More from Thursday 8 October →