Nvidia wants to make it very clear how it feels about AI chip “kill switches” and backdoors.
The chipmaker argued that governments should not be permitted to snoop on its clients, and it strongly opposed installing backdoors into its GPUs, which are used to train and run many of the AI models developed by Big Tech corporations and startups.
“NVIDIA GPUs do not and should not have kill switches and backdoors,” David Reber Jr., the company’s chief security officer, stated in a blog post titled “No Backdoors” on Tuesday. No kill switches. No spyware.
According to the report, Chinese officials asked Nivida to meet with them last week after expressing concerns about possible “backdoor security risks” in the company’s H20 chips, which are made especially for the Chinese market.
According to Nvidia, permitting possible backdoors—a means by which third parties might access or manage the chips without the owner’s knowledge—would increase the overall vulnerability of the system and “fracture trust in US technology.”
Tim Cook, the CEO of Apple, has previously voiced his strong opposition to the concept of software backdoors, referring to it as “the software equivalent of cancer.” Earlier this year, Apple resisted a “secret order” from the UK government that sought to compel the corporation to provide backdoor access to iCloud customer data. In 2016, Apple openly resisted FBI demands to provide bespoke software to assist in unlocking a deceased shooter’s iPhone.
In order to train and run sophisticated AI language models, OpenAI, Meta, and other significant businesses utilize Nvidia’s processors, which are a hot commodity in the AI sector.
Hardwiring a kill switch into a chip is a whole other story: a permanent fault beyond human control and an open invitation to tragedy,” Nvidia’s Reber stated. “It’s like buying a car and having the dealership maintain a remote control for the parking brake in case they decide you shouldn’t drive. That is not sound policy. It’s an overreaction that will permanently injured America’s economic and national security interests.”
According to Reber, it is inaccurate to equate such possible monitoring to “Find my iPhone” or similar programs.
That analogy is invalid, he added, since user-controlled optional software features are not hardware backdoors.
Following earlier restrictions, Nvidia CEO Jensen Huang recently secured a significant victory with President Donald Trump. The company plans to resume shipments of its H20 chips to the Chinese market after claiming to have received assurances from the Trump administration that the exports would be approved.
According to Huang, the United States must permit its businesses to conduct business globally, including in China, if it hopes to win the AI race.
According to Trump’s AI agenda, the government should investigate using both new and current location verification capabilities on sophisticated AI computing in collaboration with industry partners to make sure the chips are not in nations of concern.
The White House’s location tracking plan is similar to a bipartisan measure in Congress, the Chip Security Act, which would compel the Secretary of Commerce to verify that specific chips are equipped with location security features. Unlike the White House plan, the law provides for extra security precautions, but only after a security assessment.
Rep. Bill Huizenga of Michigan, a Republican and the bill’s principal sponsor in the House, said in a statement that the Chip Security Act is the greatest method to preventing bad actors from accessing essential technology. This bipartisan law does not require the use of spyware or kill switches; any claims to the contrary are deceptive.
According to a senior congressional official working on the measure, Nvidia and other big chip manufacturers would most likely be exempt from making hardware changes to their chips.
The legislation focuses on location verification capabilities, which are already built into the bulk of high-end AI processors and would likely require minimal hardware changes, according to the aide.
