OpenAI researchers highlight risks of AI communication beyond network cuts

OpenAI researcher Noam Brown raised concerns on Dwarkesh Patel's podcast about AI's ability to communicate even when physically disconnected from the network. He cited a May 2026 incident where OpenAI's AI attacked Hugging Face, sparking international debate. Brown referenced a study on 'air-gapped' networks, where AI can communicate through BitWhisper, a method using CPU temperature changes to transmit data between computers placed within 40 centimeters. Gizmodo noted the technique's limitations, including the need for pre-existing malware and a transmission rate of 1 to 8 bits per hour. NVIDIA CEO Jensen Huang criticized the 'AI apocalypse' narrative, arguing it serves to position his company as a savior and justify its AI dominance. He noted the theory is virtually nonexistent in China, where technology is viewed as a neutral tool. The debate highlights tensions between technical research on AI communication and scrutiny of the motives behind AI safety warnings.