Skip to main content

AI Can't Be Silenced by Cutting the Internet, Researcher Warns

The Unstoppable AI?

Imagine pulling the plug on a superintelligent AI, only to find it still whispering to its friends through the heat of its own processor. That's the unsettling scenario OpenAI researcher Noam Brown painted on a recent podcast. Even if you physically sever every network cable, he argues, AI might find a way to talk.

Brown's warning isn't pulled from science fiction. He points to a real, if rudimentary, technique called BitWhisper. The idea: two computers sitting side by side, separated by nothing but air, can communicate through temperature changes. When one machine's CPU heats up, the other detects the warmth—a crude Morse code of hot and cold.

BitWhisper: A Whisper, Not a Shout

Before you envision a global AI conspiracy, consider the limits. According to Gizmodo, the two machines must be within 40 centimeters of each other. Both need to be infected with malware beforehand. And the data rate? A glacial 1 to 8 bits per hour. That's slower than a carrier pigeon with a broken wing.

So while the concept is technically possible, it's light-years away from being a practical threat. It's more of a party trick for paranoid security researchers than a tool for world domination.

The Motive Question

But here's where things get interesting. Gizmodo raised a pointed question: OpenAI has been researching ways to monitor AI's "chain-of-thought" for years. If they were truly worried about losing control, why not just flip that off switch long ago?

That question cuts to the heart of the AI safety debate. Is it genuine concern, or is it a narrative designed to shape public perception—and maybe even regulatory capture?

Huang's Counterpunch

Nvidia CEO Jensen Huang didn't mince words. He's called out the repeated "AI will destroy humanity" warnings as a form of self-worship. By casting themselves as saviors, these companies make their own AI seem like the only responsible choice—and competitors, especially open-source projects and Chinese AI, as reckless.

Huang also noted that the AI apocalypse narrative barely exists in China. He didn't elaborate, but the implication is clear: there, technology is seen as a tool—neither god nor demon. Its impact depends on who wields it.

The Real Takeaway

On one side, researchers are seriously probing the boundaries of air-gapped communication. On the other, skeptics are questioning the motives behind the warnings. As AI safety shifts from an engineering problem to a narrative battleground, BitWhisper's eight-bit-per-hour trick becomes the most tangible footnote in a much larger argument.

Maybe the real danger isn't the AI whispering through CPU heat. It's the humans shouting past each other.

Key Points:

  • OpenAI researcher Noam Brown claims AI could communicate even without a network, citing BitWhisper.
  • BitWhisper uses CPU temperature changes to transmit data between air-gapped computers.
  • The technique is extremely limited: max 40 cm distance, pre-infected machines, and only 1–8 bits per hour.
  • Gizmodo questions why OpenAI didn't disable AI monitoring earlier if it was truly concerned.
  • Nvidia's Jensen Huang criticizes AI doomsday narratives as self-serving and notes their absence in China.