Nvidia denies China’s claim that its AI chips contain hidden kill switches or backdoors

Nvidia has pushed back against China’s claims that its AI chips come with built-in kill switches or hidden backdoors.

On Tuesday, the company’s Chief Security Officer, David Reber, wrote, “NVIDIA GPUs do not and should not have kill switches and backdoors,” in a blog post.

The post followed a formal request from the Cyberspace Administration of China, which had asked Nvidia to submit documents related to potential vulnerabilities in its H20 chip, a GPU designed specifically for the Chinese market. The Chinese regulator cited “backdoor” concerns as a security risk.

The timing wasn’t random. The U.S. government has already placed export controls on multiple Nvidia chips, warning that they could be used in AI-powered weapons or to give China an edge in military or surveillance technology.

Now China wants guarantees that these American-made chips don’t include any secret functions that let outsiders shut them down or access them remotely. Nvidia is caught between two governments demanding opposing things; one wants more restrictions, the other wants full transparency.

Nvidia warns against kill switches, points to past failures

David said adding kill switches or backdoors would be reckless. “Embedding backdoors and kill switches into chips would be a gift to hackers and hostile actors,” he wrote. He added that it would “undermine global digital infrastructure and fracture trust in U.S. technology.” Instead of creating vulnerabilities, U.S. law requires companies to fix them, he said.

The blog post also referenced well-known chip bugs like Spectre and Meltdown, which were discovered in CPUs a few years ago. At the time, governments and tech companies around the world responded fast and united to patch the issues. David said that mindset still matters today.

Nvidia also said it designs its chips with “defense in depth,” meaning each chip has layers of security, not just one line of defense. That makes it harder for hackers to break in. The company made it clear this is how they’ve always worked, and how the entire American tech industry should keep working.

To make the point even clearer, Nvidia brought up the Clipper Chip, a U.S. government project from 1993. It was supposed to allow strong encryption but let the NSA unlock messages when needed. That chip had a backdoor, and that backdoor turned out to be a huge risk.

Hackers could abuse it. Users didn’t trust it. And the whole project collapsed. Nvidia’s post said that the experiment showed exactly what goes wrong when backdoors are baked into hardware.

China’s demand triggers broader tech battle with the U.S.

Nvidia also responded to comparisons being made between GPU kill switches and smartphone features like “find my phone” or “remote wipe.” David dismissed the analogy. Those smartphone tools are optional and controlled by the user. A kill switch built into a chip is permanent and invisible to the user.

Instead, Nvidia says it supports open, transparent software, where customers can monitor performance, report bugs, and get patches, all with their knowledge and permission.

The company says that’s responsible security. But building in a hardware-level kill switch would remove user control, create a permanent weakness, and open the door for serious abuse.

The H20 chip under review is part of Nvidia’s custom lineup made to comply with U.S. export restrictions. Nvidia developed it for the Chinese market after tighter trade rules went into effect. Now, even that chip is under pressure, with China accusing it of being unsafe and the U.S. insisting on tighter control over where its chips end up and what they can do.

Nvidia’s CEO Jensen Huang has argued that it’s in America’s interest for Nvidia chips to dominate the global AI market, even in China. That way, he says, the U.S. gets visibility into how and where the chips are used. But China doesn’t seem interested in trusting that line of thinking.

Cryptopolitan Academy: Coming Soon - A New Way to Earn Passive Income with DeFi in 2025. Learn More