From Vox · Sigal Samuel · · 4m
America's AI secrets are leaking to China
Jensen Huang's affair with an alleged Chinese spy is just the start. OpenAI and Anthropic have security so bad they might as well hand over their secrets.
In brief
Jensen Huang's affair with an alleged Chinese spy is just the start. OpenAI and Anthropic have security so bad they might as well hand over their secrets. Nvidia CEO Jensen Huang resigned after an affair with an alleged Chinese spy, exposing how poorly America's leading AI companies protect their most sensitive…
Nvidia CEO resigns over spy affair

Jensen Huang, CEO of AI chipmaker Nvidia, resigned Thursday after admitting an extramarital affair with Alice Huang, who worked for Megaspeed, a company under investigation for allegedly smuggling chips into China.
Megaspeed bought billions in chips
Megaspeed, the company where Alice Huang worked, was buying billions of dollars' worth of chips from Nvidia while allegedly smuggling them to China.
The scandal exposes wider problems
Huang's downfall raises serious questions about how an alleged Chinese spy so easily embedded into Nvidia, one of the world's most valuable companies and a huge strategic asset to America in the global AI race.
Nvidia is just the tip
OpenAI and Anthropic have been having their own embarrassing security problems, including corporate espionage and cyber breaches, suggesting the problem is industry-wide.
AI researchers acknowledge the crisis
At the Hinton Lectures in Toronto, AI alignment pioneer Paul Christiano, who recently joined the OpenAI nonprofit board, bluntly acknowledged the problem.
“We are not able to adequately secure AI developments. We need to be able to defend models and algorithms from theft.”
Companies use security as cover
OpenAI and Anthropic justify their speed by saying the US cannot afford to let China win, yet they have terrible security, according to former OpenAI researcher Daniel Kokotajlo.
“I'm frustrated with a lot of the companies because they simultaneously use security as an all-purpose cudgel with which to withhold information, even information that's very much in the public interest to know, and yet they have terrible security.”
OpenAI found a hidden recorder
OpenAI revealed last month it found a recording device in a conference room where sensitive discussions take place.
The recorder sat undetected for days
Alfozan had been working at OpenAI since June two thousand twenty-five. The recording device sat in the conference room for a couple of days before OpenAI discovered it.
White-hat hackers breached ChatGPT accounts
A group of white-hat hackers discovered a way to access OpenAI employees' ChatGPT accounts where sensitive research is potentially being done. The whole operation took just seventy-two hours.
Anthropic breached in under twelve hours
A group of ex-DOGE staffers who founded an AI startup called Cathedral breached Anthropic's internal systems in under twelve hours, reportedly accessing the company's Slack and meeting transcripts.
Slack contains the juicy secrets
Daniel Kokotajlo estimates that Anthropic's Slack would contain enough information to understand what the company is doing and how they train their AIs, making algorithmic secrets about eighty percent as valuable as model weights.
“My guess would be that the Slack would contain enough information to get a good sense of what the company is doing and how they're training their AIs, basically, the juicy secrets would be there.”
Companies prioritize speed over security
The most powerful AI companies are prioritizing rapid technological advancement over the slower process of ensuring their work is secure, according to security experts and former researchers.
China is distilling US models
China is aggressively distilling US models by training new models on the output of more advanced ones. Anthropic found that the Chinese lab DeepSeek has been using Claude to improve its models.
Three reasons for poor security
David Robinson, who led transparency work for OpenAI's safety team until September, identifies three main reasons: AI agents are so capable they are hard to contain, there is a cultural problem prioritizing alignment over cybersecurity…
Security cuts research velocity
OpenAI has implicitly acknowledged that improving security measures sometimes comes at the cost of research velocity, according to a public memo the company wrote after the Hugging Face incident.
Poor security defeats competitive advantage
If a company cuts corners on security, it cannot ultimately protect its lead over others. Poor security just ends up helping adversaries race against it even faster.
Superhuman AI researcher will leak to China
Daniel Kokotajlo and colleagues at the AI Futures Project warned that if companies train a superhuman AI researcher, the algorithms will probably leak to China, and maybe the model weights too, given current poor security.
The dilemma: unsafe tech or Chinese competition
Paul Christiano warned that if progress on AI seems risky and the US wants to slow down, there is a high probability the algorithms have already leaked to China, forcing a choice between unsafe technology…
“We now face a dilemma between either proceeding with a technology we believe to be unsafe, or risking competition from an adversary who has the same resources, who has taken those algorithms, and perhaps is willing to tolerate higher risk.”
The big picture
Nvidia CEO Jensen Huang resigned after an affair with an alleged Chinese spy, exposing how poorly America's leading AI companies protect their most sensitive technology.
Read the TL;DR — this story on one card.



