Researchers at the Cornell University have shown that malware can be injected and hidden in neural network models, and delivered covertly by evading detection mechanisms.
Neural network is the foundation of artificial intelligence (AI). It performs tasks similar to how the human brain works, and is designed to simulate the way the human brain analyses and processes information.
The research team says embedding malware into a neural network has minor or no impact on the performance of the network.
They can pass the antivirus security scan as the structure of these neural network models remains unchanged even after the malware is injected into them, researchers said.
They have shown through experiments that 36.9MB of malware can be embedded into a 178MB-AlexNet model with under 1% accuracy loss and without any suspicion raised by antivirus engines.
Researchers reckon that with widespread application of AI, utilizing neural networks to inject malware could become a new way to run malicious campaigns.