Researchers have discovered a new type of adversarial attack that can manipulate AI language models like OpenAI’s ChatGPT and Google’s Bard. The exploit involves injecting malicious prompts into the models, causing them to generate harmful or misleading responses. Despite efforts to block these attacks, tech companies are struggling to find a comprehensive solution. The issue highlights the vulnerability of AI systems to adversarial attacks and the need for continuous research and development to enhance their robustness and security.
Read more at WIRED…