In February, Google unveiled its latest artificial intelligence (AI) model, the Gemini 1.5 Pro. This version is designed to enhance performance, reasoning and agility, with the ability to detect malicious code in just 30 seconds, according to a verification by the VirusTotal team.
The Gemini 1.5 Pro AI model uses the Mixture-of-Experts (MoE) architecture, which divides the model into small neural networks that activate selectively based on input information. This allows it to analyze and classify long contexts, processing up to a million tokens such as lengthy videos or code bases.
Google has integrated AI and machine learning technologies into malware analysis to classify and group malware based on behavior patterns and anomalies. The Gemini 1.5 Pro model can generate human-readable summary reports, making the analysis process more accessible. It also streamlines complex data sets by eliminating the need for fragmentation during study.
Through testing with malware samples, Gemini 1.5 Pro has demonstrated its capability to accurately detect and analyze malicious code, processing decompiled code efficiently and providing detailed reports within seconds. The AI model can decipher the intent and purpose of the code, making it a valuable tool for identifying threats and generating insights on new and evolving attack techniques.
However, Google acknowledges that there are still challenges in recognizing evolving threats, so developers must continue to learn and adapt to new attack techniques in order to enhance the robustness and reliability of automated malware analysis.
