Particle.news

Download on the App Store

Nvidia Rejects Kill Switches and Backdoors in AI Chips

Nvidia warns that hardware kill switches or backdoors would create dangerous vulnerabilities, fracturing trust in global digital infrastructure.

BEIJING, CHINA - AUGUST 01: In this photo illustration, an Nvidia chip is seen through a magnifying glass on August 1, 2025 in Beijing, China. (Photo by VCG/VCG via Getty Images)
Image
Image
Michael Kratsios during the Bloomberg Tech conference in San Francisco, California, US, on Thursday, June 5, 2025.

Overview

  • Nvidia’s Chief Security Officer published a bilingual blog post categorically denying any kill switches or backdoors in its GPUs and urging U.S. lawmakers to abandon proposals for hardware-level controls.
  • Bipartisan bills in Congress and White House recommendations are still under review to require location verification and potential shutdown features in exported advanced AI chips.
  • China’s Cyberspace Administration summoned Nvidia representatives over alleged spyware and remote-shutdown risks in the H20 chips, prompting the firm’s public rebuttal.
  • Nvidia and security specialists argue that embedding permanent backdoors or kill switches would introduce single points of failure, inviting exploitation by hackers and hostile actors.
  • The debate underscores intensifying U.S.–China competition for AI hardware dominance and the rapid evolution of export-control policies on semiconductors.