Researchers have discovered a new method of slipping malware past automated detention tools by hiding it in a neutral network.
To prove the validity of the technique, the researchers embellished 36.9 MiB of malware in a 178 MiB AlexaNet model without significantly altering the function of the model itself.
The malware-embedded model classified images with almost identical accuracy within 1% of the malware-free model.
By selecting the best layer to work with in an already trained model, and then embedding the malware in that layer, the researchers were able to break the malware in a way that allowed them to bypass detection by standard antivirus engines.
The new technique is a way to hide malware, not execute it. To actually execute the malware, it must be extracted from the poisoned model by another malicious program and then reassembled into its working form.
Researchers Zhi Wang, Chaoge Liu, and Xiang Cui made the discovery.
For more information, read the original story in Arstechnica.