This article discusses GTIG’s approach to understanding and countering the activities of government-backed threat actors, particularly their use of AI tools like Gemini. The analysis highlights the methods used by these actors, their experimentation with AI for operational efficiency, and the limitations they face in leveraging such technologies for malicious purposes. Overall, while AI assists in speeding up common cyber-attack processes, it has not yet enabled breakthroughs in adversarial capabilities. Affected: Google platforms, users, government-backed threat actors, Iranian APT actors, Russian APT actors, IO actors.
Keypoints :
- GTIG employs an intelligence-driven approach to detect and disrupt government-backed threat activities.
- Threat actors are using AI tools like Gemini primarily for routine tasks and not for novel capabilities.
- Generative AI is being used by threat actors for research, troubleshooting, and content generation.
- Most notable users of Gemini include Iranian APT actors, with limited participation from Russian APT actors.
- AI tools currently serve to enhance productivity but do not significantly change the threat landscape.
- Threat actors attempted to abuse Gemini to bypass safety controls, but most efforts were unsuccessful.
- Low-effort jailbreak attempts were made using publicly available prompts, which did not yield malicious capabilities.
MITRE Techniques :
- Tactic: Initial Access – Technique: Phishing (T1566) – Threat actors researched Gmail phishing techniques to exploit Google accounts.
- Tactic: Reconnaissance – Technique: Search Open Websites/Domains (T1593) – Used Gemini for reconnaissance on target organizations.
- Tactic: Defense Evasion – Technique: Masquerading (T1036) – Attempts were made to bypass Gemini’s safety controls.
- Tactic: Command and Control – Technique: Application Layer Protocol (T1071) – Assistance with malicious scripting and evasion techniques.
- Tactic: Collection – Technique: Data from Information Repositories (T1213) – Attempted to find sensitive user data and exfiltrate it.
Full Story: https://cloud.google.com/blog/topics/threat-intelligence/adversarial-misuse-generative-ai/