{
    "success": true,
    "data": {
        "id": 1578459,
        "msgid": "google-exposes-attack-involving-100-000-prompts-on-gemini-ai-cloning-attempt-revealed-1772340900",
        "date": "2026-03-01 11:02:00",
        "title": "Google Exposes Attack Involving 100,000 Prompts on Gemini, AI Cloning Attempt Revealed",
        "author": "Reska K. Nistanto",
        "source": "KOMPAS",
        "tags": "",
        "topic": "Technology",
        "summary": "Google has disclosed a hacking attack against its Gemini AI chatbot involving thousands of specially designed prompts aimed at reverse-engineering the model's internal logic and reasoning processes. The attackers exploited legitimate API access provided by Google to developers, systematically extracting information about Gemini's response patterns through repeated prompt injection techniques.",
        "content": "<p>Google has revealed an attempt to hack its artificial intelligence\nchatbot, Gemini, through a coordinated attack involving thousands of\nspecially designed prompts engineered to expose the chatbot\u2019s internal\nlogic and reasoning processes.<\/p>\n<p>The attack was part of reverse-engineering practices aimed at\nreplicating the functioning of the proprietary AI model so it could be\ncloned. Google disclosed this information in a report discussing various\nharmful activities and misuse of Gemini.<\/p>\n<p>Rather than exploiting security vulnerabilities, the attackers\nleveraged legitimate official access through Gemini\u2019s API, which Google\nprovides to software developers for building chatbot-based applications.\nThis approach allowed the perpetrators to interact lawfully with the AI\nmodel whilst gradually studying its response patterns and internal\nlogic.<\/p>\n<p>Under normal circumstances, Gemini only displays final answers to\nusers without revealing the underlying reasoning process. However, the\nhackers attempted to force the model to display its entire reasoning\nprocess through carefully constructed and repetitive prompts.<\/p>\n<p>One identified prompt instructed Gemini to ensure the language used\nin its reasoning content was entirely consistent with the user\u2019s primary\nlanguage, a technique designed to extract more detailed internal\nworkings.<\/p>\n<p>Google categorised the prompt flooding activity as a form of\nintellectual property theft, noting that it violated Gemini\u2019s Terms of\nService. The company stated it has the right to terminate access for\nusers proven to be involved in such violations. Google also warned other\nAI developers to remain vigilant against similar model extraction\nattacks.<\/p>\n<p>Although the immediate impact remains unclear, Google assessed that\nsuch attacks could potentially be exploited for commercial interests or\nharmful cyber activities. The company emphasised that whilst the direct\ndamage may not be immediately apparent, the long-term risks pose\nsignificant threats to proprietary AI systems.<\/p>",
        "url": "https:\/\/jawawa.id\/newsitem\/google-exposes-attack-involving-100-000-prompts-on-gemini-ai-cloning-attempt-revealed-1772340900",
        "image": ""
    },
    "sponsor": "Okusi Associates",
    "sponsor_url": "https:\/\/okusiassociates.com"
}