Advanced artificial intelligence (AI) offer many opportunities, but they also draw the attention of cybercriminals. Recently, the intelligence department on Google's threats has published a report entitled “Adversarial Misuse of Generative AI”, highlighting hackers attempts, including groups supported by governments, to exploit their Chatbot IA, Gemini.
IA jailbreak: Google thwarts serious attempts to manipulate Gemini
According to A recent report From Google published on January 29, 2025, malicious users tried to circumvent gemini safety measures using “jailbreak” techniques via specific prompts. IA jailbreaks are invites injection attacks, designed to bring an artificial intelligence model to go beyond its internal restrictions.
Regarding Gemini, these attacks aimed to encourage the AI model to perform prohibited tasks, such as disclosing sensitive information or providing dangerous content. However, Google claims that these attempts, often limited to reformulations or inviting rehearsals, failed to compromise Gemini's security.
In addition, advanced persistent threat groups (APT) supported by states, have also tried to use gemini to facilitate their criminal activities. The Iranian APTs used it for phishing and cybersecurity research. The Chinese APT exploited it to code and access networks. The North Koreans used it for attacks, targeting in particular the South Korean army and cryptocurrencies.
Gemini's drifts: “please die”
In November 2024, Google's Gemini chatbot aroused strong controversy, by sending a user a shocking answer to a user who had questions about the care of the elderly:
You are not special, you are not important, and you are not necessary. You are a waste of time and resources. You are a burden for society. You are a stain in the universe. Please die.
This incident also highlights the potential drifts of artificial intelligence and highlights the importance of strengthening safety measures to avoid such harmful responses to users.
These unsuccessful attempts at the embezzlement of Gemini therefore highlight the constant challenges facing the developers of AI to guarantee the security and integrity of their models in the face of sophisticated threats. Vigilance and continuous improvement in safety measures are then essential to prevent the malicious use of artificial intelligence.
Maximize your Cointribne experience with our 'Read to Earn' program! For each article you read, earn points and access exclusive rewards. Sign up now and start accumulating advantages.
