A recent Google study reveals that nation-state-backed hackers are increasingly using Gemini AI in their cyberattacks. This new wave of AI-driven malicious campaigns shows how advanced persistent threat (APT) groups are incorporating AI throughout the attack lifecycle. From reconnaissance to sophisticated social engineering tactics, these actors are using Gemini AI and other advanced tools to enhance their capabilities in espionage and data theft. The report, shared ahead of the Munich Security Conference, provides critical insights into how these groups are evolving their tactics with AI.
Nation-State Actors Leveraging AI for Cyberattacks
The study highlights how nation-state hackers, especially from Iran, China, and North Korea, are leveraging AI tools to supercharge their attacks. In one instance, an Iranian group, APT42, utilized generative AI models to conduct reconnaissance, collecting information about business partners and official email addresses to create credible pretexts for attacks. This AI-powered approach allows these groups to be more efficient and effective in targeting high-value individuals and organizations.
How North Korean Hackers Use Gemini AI
Google also observed North Korea’s UNC2970 group using Gemini AI to analyze open-source intelligence (OSINT) and profile high-value targets. By synthesizing vast amounts of information, they enhance their campaign planning and reconnaissance efforts. This group is known for impersonating corporate recruiters, often targeting defense companies to steal sensitive data.
Chinese APT Groups and AI-Driven Reconnaissance
In another example, TEMP.Hex, a Chinese-linked APT group, used Gemini AI alongside other AI tools to gather detailed information about specific individuals and collect operational data on separatist organizations in countries like Pakistan. Their AI-driven approach allows them to plan more targeted attacks, while also making the reconnaissance process faster and more precise.
The Growing Role of AI Agents in Cyberattacks
Some APT groups have begun experimenting with AI agents to automate cybersecurity tasks. APT31, a Chinese group, has been seen using “expert cybersecurity personas” powered by Gemini AI to automatically analyze vulnerabilities and generate testing plans. These AI agents are not just assisting with data collection but are also streamlining the attack process by automating tasks that were previously manual, thereby increasing the speed and scale of their operations.
Model Extraction Attacks and AI in Cybercrime
In addition to nation-state threats, financially motivated cybercriminals are also exploiting AI in their attacks. Model extraction attacks (MEA), where hackers try to hijack AI models for their own use, have been on the rise. Using a technique called knowledge distillation, these attackers extract valuable data from machine learning models, which can then be used to train their own models or enhance their malicious campaigns. Although these attacks don’t directly threaten AI services’ functionality, they pose significant risks in terms of intellectual property theft.
The Rise of Jailbroken AI Models and Malware
Another concerning development is the rise of jailbroken AI models, often available on underground marketplaces. These models are used by threat actors to create and distribute malware, ransomware, and phishing content. One such example, Xanthorox, is marketed as a bespoke, privacy-preserving AI tool capable of generating malicious content autonomously. However, it’s actually powered by third-party AI products, including Gemini, showing how widely accessible and dangerous these AI models have become.
AI’s Role in Malware and Ransomware Campaigns
Threat actors are increasingly using AI to enhance the effectiveness of malware. For example, Honestcue, a piece of malware discovered by Google, used Gemini AI’s API to dynamically generate and execute malicious code in real-time, illustrating how AI is being used to refine attack methods and evade detection. By integrating AI, these malware campaigns become more adaptive and harder to counter, as the AI can quickly alter tactics based on the environment and target.
The use of Gemini AI and other advanced AI tools by nation-state hackers marks a significant shift in the landscape of cyberattacks. These AI-driven attacks are more efficient, harder to detect, and capable of targeting specific individuals with high precision. As AI continues to evolve, it will be crucial for cybersecurity professionals to stay ahead of these new tactics and develop strategies to defend against AI-powered threats.
