Dec 27, 2024
AI’s Achilles’ Heel: Researchers Expose Major Model Security Flaw
Posted by Genevieve Klien in categories: robotics/AI, security
Researchers used electromagnetic signals to steal and replicate AI models from a Google Edge TPU with 99.91% accuracy, exposing significant vulnerabilities in AI systems and calling for urgent protective measures.
Researchers have shown that it’s possible to steal an artificial intelligence (AI) model without directly hacking the device it runs on. This innovative technique requires no prior knowledge of the software or architecture supporting the AI, making it a significant advancement in model extraction methods.
“AI models are valuable, we don’t want people to steal them,” says Aydin Aysu, co-author of a paper on the work and an associate professor of electrical and computer engineering at North Carolina State University. “Building a model is expensive and requires significant computing sources. But just as importantly, when a model is leaked, or stolen, the model also becomes more vulnerable to attacks – because third parties can study the model and identify any weaknesses.”