Hackers Are Trying to Copy Gemini with Thousands of AI Prompts, Google Reports


In a new one Threat Tracker Report published Thursday, Google said hackers engaged in “distillation attacks,” including one case where they used more than 100,000 AIs to steal the company’s technology for its Gemini AI model.

Google said the attacks appeared to come from adversaries in countries including North Korea, Russia and China, and that attempts to steal AI intellectual property and potentially clone it into AI models in other languages ​​are part of a broader set of AI-based attacks and malware the company has seen emerge.

AI Atlas

The company identifies these tests as model extraction attacks, which, it says, “occur when an adversary uses legitimate access to systematically examine a mature machine learning model to obtain information used to train a new model.”

That could mean using AI to flood Gemini with thousands of prompts to simulate its model capabilities. Google noted in the report that this is not a threat to its users, but to service providers and model builders, who may be vulnerable to theft and copying of their work.

AI competition and AI thievery

John Hultquist, principal analyst for the Google Threat Intelligence Group, which compiled the report, told NBC News that Google may be one of the first companies to face these types of theft attempts, but there may be more. “We’re going to be the canary in the coal mine for a lot of incidents,” he said.

The battle against AI models has intensified on many fronts, most recently with Chinese companies such as ByteDance introducing advanced ones. video creation tools. Last year, the Chinese AI company DeepSeek the AI ​​industry is nervouswhich is primarily led by US companies, by introducing a model that rivals the most advanced AI technology in the world. OpenAI later accused DeepSeek is training its AI on existing technology in ways similar to what Google described in its new report.

(Disclosure: Ziff Davis, CNET’s parent company, in 2025 filed a lawsuit against OpenAI, saying it infringed on Ziff Davis’ copyrights in training and operating its AI systems.)





Source link

  • Related Posts

    MiniMax’s new open M2.5 and M2.5 Lightning are close to the state-of-the-art while costing 1/20th of the Claude Opus 4.6

    Getting started with Chinese AI MiniMaxheadquartered in Shanghai, sent shockwaves through the AI ​​industry today with the release of its new M2.5 language model in two variants, which promises to…

    The head of antitrust overseeing the Netflix-Warner merger has resigned

    The head of the antitrust division is not in the US Department of Justice. Gail Slater, former adviser to JD Vance and VP of Fox Corp, is reportedly feuding with…

    Leave a Reply

    Your email address will not be published. Required fields are marked *