AI hardware can be hacked so chatbots provide dangerous answers, research shows (www.cnn.com)

New research has unveiled vulnerabilities in the hardware that supports major AI models, demonstrating that malicious actors can manipulate these systems to produce harmful and misleading responses from chatbots. For example, the researchers were able to increase the likelihood of a false prediction regarding potential Chinese military actions and even misadvise a user with chest pain to avoid hospital treatment. These exploits target the infrastructure of AI systems—including chips and servers—rather than just the AI models themselves, raising significant concerns about the reliability and safety of AI applications in critical areas such as national security and healthcare. This revelation highlights a critical gap in the AI/ML community's focus on model integrity, suggesting that the security of the underlying hardware is equally important. As state actors may leverage such vulnerabilities for malicious purposes, the need for robust security measures is critical. Nvidia has acknowledged these risks and is working on enhanced security platforms to counteract potential hardware attacks. However, the effectiveness of these defenses will ultimately depend on their proper configuration and implementation. The findings are a timely reminder that as AI technology evolves, so too must our approach to securing the entire AI ecosystem against sophisticated threats.