The advancement of artificial intelligence opens new opportunities, but also increases the risk of misuse. Google's report reveals details of such threats concerning the Gemini AI bot.
Unsuccessful Attempts to Breach Gemini
Google reports that while attempts were made to breach the Gemini AI bot, no advanced attacks were identified. Hackers used tactics like rephrasing prompts and repeatedly sending them, but all attempts were thwarted. AI jailbreaks are often aimed at executing prohibited tasks such as leaking sensitive information and providing unsafe content.
Use of Gemini by Government-Backed APT Groups
Google reports that government-backed APT groups attempted to utilize Gemini for malicious purposes such as intelligence gathering, researching vulnerabilities, and coding tasks. These also included post-compromise activities like defense evasion. For instance, Iranian groups used AI for phishing campaigns and generating cybersecurity content.
Report on Cyber Attacks by APT Groups from Iran, China, and North Korea
Chinese groups reportedly used Gemini for code troubleshooting and exploring network access methods. North Korean APT groups applied AI in all stages of their attack lifecycle, including research and development. According to Chainalysis, North Korean hackers stole $1.3 billion in 2024.
Increased efforts by hackers to exploit AI for malicious ends necessitate advanced defense measures and innovative approaches to AI security. Google's reports highlight the significance of these threats and the need for vigilance.
