Cybersecurity experts have developed a novel software capable of breaching any artificial intelligence system
A team of international researchers has unveiled a distinctive network worm with the capability to target services powered by generative artificial intelligence, including OpenAI’s ChatGPT and Google’s Gemini. This malicious software is not only capable of independently spreading across AI services but also capable of data theft and spam distribution.
The malicious software, named Morris II, is engineered to launch assaults on AI-based virtual assistants, effectively bypassing current protective measures. The worm leverages the large language models’ ability to process text, images, and video content.
The researchers demonstrated how an email attack could compel an AI to seek additional information online, leading to a “breach” and the theft of confidential data from emails.
To combat such attacks, experts recommend not only enhancing the reliability of AI systems but also revising their usage approach. It’s crucial for users to limit the privileges granted to AI, such as prohibiting them from sending emails on their behalf. Additionally, it’s important that all system actions are aligned with human oversight.
